At Microsoft, we’re developing a growing agent ecosystem that offers diverse options for your workplace demands. This collection of tools acknowledges that there is no single “right” way to build an agent.

“The best agent architecture isn’t the most powerful one. It’s the simplest one that delivers value today while giving you the flexibility to grow tomorrow.”

Patrice Pelland, partner engineering manager, HR Employee Experience

Microsoft 365 Copilot, declarative agents, Copilot Studio, Microsoft Fabric data agents, and Microsoft Foundry agent service all represent valid patterns that are optimized for different kinds of work. Newer ideas like scout-style personal agents and Frontier Tuning (how we fine-tune our agent compliance) offer even more flexibility.

When choosing the best agentic technology for you, the rule of thumb is simple: Select the lightest architecture that fits your user experience, data needs, action model, and governance requirements.

“The best agent architecture isn’t the most powerful one,” says Patrice Pelland, a partner engineering manager in HR Employee Experience. “It’s the simplest one that delivers value today, while giving you the flexibility to grow tomorrow.”

Start with the surface, not the stack

The fastest way to get agent architecture wrong is to begin with technology instead of the workload. A better starting point is the user surface—where the work happens and how people expect to interact with it.

Our agent ecosystem is broad, but each of the core options map cleanly to a different kind of challenge:

• Microsoft 365 Copilot is the default entry point for AI in daily work, embedded across apps like Teams, Outlook, and Word. It excels at summarization, drafting, and reasoning over Microsoft Graph data without requiring users to leave their most familiar tools.
• Microsoft Copilot Cowork takes Microsoft 365 Copilot beyond just answering questions to actually executing on work assignments. Designed for complex, multi-step tasks, it can plan, coordinate, and carry out work across Microsoft 365 applications and connected systems while keeping users in control through approval checkpoints and oversight.
• Declarative agents extend Microsoft 365 Copilot with scoped instructions, knowledge, and lightweight actions. They’re the simplest way to create focused, role-based assistants while staying entirely within the Copilot experience.
• Copilot Studio agents provide a low-code environment for building workflow-driven agents with connectors, triggers, approvals, and multi-channel delivery. They’re best suited for business processes that need structure, governance, and integration across systems.
• Microsoft Foundry Agent Service is a code-first platform for building and operating custom agents with full control over models, orchestration, and runtime. It fits engineering-heavy scenarios that require scalability and tailored architecture.
• Microsoft Fabric data agents specialize in governed analytics over business data, using semantic models, ontologies, and curated datasets. They are designed to answer questions about KPIs, entities, and operations with trusted, auditable context rather than general document retrieval.
• Microsoft Scout and Autopilots introduce always-on personal agents that can coordinate work in the background across Microsoft 365. They focus on continuity—tracking commitments, preparing context, and proactively surfacing what matters next.
• Frontier Tuning enables organizations to teach agents how work actually gets done using company-specific data, workflows, and evaluation signals. It’s a later-stage capability that refines behavior inside compliance boundaries, rather than serving as a starting point.

Taken together, these are layers in a system. The practical decision is to pick the one closest to the user.

Picking the right agentic tool

Use Microsoft 365 Copilot as the default front door for broad productivity and Microsoft 365-grounded experiences. Extend it with declarative agents, connectors, plugins, or APIs when a scenario needs more domain-specific knowledge or actions. Do not treat plain Copilot as the primary architecture when the real need is a departmental specialist, governed workflow, analytics specialist, custom channel, or custom runtime. Those needs fit Copilot Studio, Fabric, or Foundry patterns better.

Use Microsoft Cowork when you want to complete a meaningful piece of work across multiple steps, files, applications, and systems. Cowork is best suited for long-running tasks that require planning, orchestration, and execution over time, while remaining grounded in Microsoft 365 context. Do not use Cowork as a replacement for workflow platforms, analytics specialists, or custom agent architectures. When a task requires governed business processes, specialized analytics, or deep runtime customization, Cowork should coordinate with other agentic tools rather than replace them.

Use Copilot Studio when governed workflow, connectors, approvals, and channel reach matter as much as the chat experience. Do not use it as the default for a small Microsoft 365 Copilot extension or for engineering-first scenarios that require custom model choice, runtime hosting, private networking, or deep orchestration; those fit declarative agents or Foundry better.

Use Microsoft Foundry when low-code or Microsoft 365-native layers hit real limits around orchestration, runtime control, model selection, deployment architecture, or engineering operations. Do not start here for a simple business workflow, no/low-code service agent, or focused Microsoft 365 assistant unless custom runtime control is truly the differentiator.

Use Fabric data agents when you’re looking for analytics expertise in a broader architecture. Fabric IQ, semantic models, and ontology can help translate raw data into business meaning. Do not use them as a general enterprise assistant or workflow engine: they are read-only, analytics-focused, and not intended for broad action-taking, unstructured document help, or custom model/runtime control.

Use Microsoft Scout when you need a personal front-door and coordination layer. It should not serve as a replacement for every enterprise agent architecture. When work crosses into governed process, business analytics, custom orchestration, or domain systems, Scout should delegate to Copilot Studio, Fabric data agents, MCP tools, or Foundry specialists.

Use Frontier Tuning only after the basics are in place: clear workflows, governed data, meaningful evaluation signals, and operational maturity. It is a later-stage optimization for institutional know-how, not the first step for a lightweight assistant or a substitute for sound architecture, data governance, or human-in-the-loop controls.

AI-ready data: The foundation of effective agents

The success of an AI agent depends less on the model it uses and more on the quality of the data that grounds it. Agents can only reason effectively when they have access to information that is trusted, discoverable, secure, and connected to the broader decision-making context. Organizations that invest in AI-ready data create the foundation for more reliable and useful AI experiences.

“The competitive advantage in the age of AI won’t come from access to models alone. Organizations also need to prioritize AI-ready data, which allows agents to reason accurately, securely, and at enterprise scale.”

Ravi Goripathy, principal software engineering manager, HR Employee Experience

AI-ready data goes beyond standard measures of data quality, such as accuracy, completeness, and consistency. It also includes the semantics, governance, and contextual information that help AI systems understand what the data means, how it should be used, and the way it relates to other information across the business. High-quality data is essential, but AI systems need data that is also meaningful, connected, and trustworthy.

As organizations deploy more copilots and agents, data readiness increasingly becomes the limiting factor for success, rather than model availability. A strong AI-ready data foundation helps agents produce better results, apply business context consistently, and operate within appropriate security and governance boundaries. The most effective AI strategies treat data readiness as a core architectural requirement.

“The competitive advantage in the age of AI won’t come from access to models alone,” says Ravi Goripathy, a principal software engineering manager in HR Employee Experience. “Organizations also need to prioritize AI-ready data, which allows agents to reason accurately, securely, and at enterprise scale.”

Building reliable agent systems

As soon as an agent moves from answering questions to taking actions, architecture matters more.

A useful pattern is to separate responsibilities into layers:

MCP tools play a key role here by exposing actions as controlled, reusable capabilities. Instead of giving an agent broad access to systems, you expose bounded functions with defined inputs, permissions, and telemetry. This improves reliability, governance, and auditability.

Similarly, not every interaction belongs in free-form chat. Tasks that require review, approval, or comparison—like approving a budget or evaluating candidates—benefit from structured UI components. MCP apps let you surface those experiences inside conversational flows without rebuilding entire applications. Rather than sending users to a separate system to complete a task, MCP apps can render forms, grids, dashboards, and other transactional experiences directly within the conversation, enabling users to review information, approve requests, update records, and take action in context.

As solutions mature, new patterns emerge. Always-on personal agents—like Scout-style Autopilots—can monitor commitments, prepare context, and proactively surface issues. In parallel, Frontier Tuning introduces a way to teach agents how work actually gets done in a specific organization, using real workflows, evaluation signals, and internal knowledge.

These are powerful capabilities, but they’re not starting points. They’re extensions of an architecture that already works.

A layered approach beats a ‘super agent’

It’s tempting to imagine a single agent that handles context, analytics, workflows, and actions. In practice, a more durable approach is a set of loosely connected specialists that each focus on a specific responsibility.

A Microsoft 365–native assistant can serve as the front door for user interaction, delegating analytics to Fabric data agents, actions to Copilot Studio workflows or MCP tools, and more complex orchestration to Foundry-based agents when needed. This separation makes systems easier to govern, test, and evolve over time.

Before moving to production, architecture should include clear ownership, permission design, governance controls, human approval for sensitive actions, evaluation processes, and monitoring. These are all essential aspects of building an agent system that can be trusted.

Deciding where to start

When faced with multiple valid options, a simple decision path helps:

• If a packaged Microsoft experience already meets most of the need, use it.
• If the experience should live in Microsoft 365, start with Copilot and a declarative agent.
• If the core problem is workflow—approvals, routing, integration—lean toward Copilot Studio.
• If the value depends on governed business data, bring in Fabric data agents early.
• If you need deep control over orchestration, models, or runtime behavior, step up to Foundry.

And across all of these: add structure only where the workload demands it. Use MCP tools for actions, structured UI where chat breaks down, and human review where risk increases.

Most importantly, resist the urge to start at the most complex layer. Many successful solutions begin as simple Microsoft 365 extensions and evolve incrementally into multi-layered systems.

Looking ahead: From tools to systems of work

The trajectory of agent architecture is clear. We’re moving from isolated assistants toward interconnected systems that combine personal productivity, business data, workflows, and continuous coordination.

“Agentic AI is shifting from isolated capabilities to connected systems of work. The organizations that lead will be those that design for evolution—starting simple, learning continuously, and scaling with intention.”

Ravi Goripathy, principal software engineering manager, HR Employee Experience

In the near term, that means more layering: personal agents coordinating work, workflow agents handling processes, analytics agents grounding decisions, and platform agents orchestrating across systems. Over time, capabilities like Frontier Tuning will make these systems more adaptive, learning from how organizations actually operate while staying within governance boundaries.

But the core principle is unlikely to change. The most successful teams won’t be the ones who adopt the most advanced architecture first. They’ll be the ones who start with the smallest viable solution, validate it in real workflows, and scale intentionally—adding complexity only when it delivers clear value.

“Agentic AI is shifting from isolated capabilities to connected systems of work,” Goripathy says. “The organizations that lead will be those that design for evolution—starting simple, learning continuously, and scaling with intention.”

The post Start light, scale intentionally: Choosing the right Microsoft agent architecture appeared first on Inside Track.

That was the guiding mantra behind an ambitious cross-company effort involving our team in Microsoft Digital—the company’s IT organization—and a number of our product teams. The effort, internally referred to as the Securing AI Agents initiative, brought together teams from Microsoft Digital, Windows, Entra, Intune, Defender, Purview, and Microsoft Security to validate secure AI agent scenarios inside Microsoft’s corporate tenant.

Together, we set out to prove that AI agents could operate safely inside a real enterprise environment, not just in a controlled demonstration.

“Securing AI in the enterprise at pace requires an integrated, full-stack approach. By validating these capabilities together at enterprise scale, we’re generating the real-world learning to strengthen Microsoft’s products and give customers a trusted blueprint for secure AI adoption.”

Ragini Singh, partner group engineering manager, Microsoft Digital

As part of our role as the company’s Customer Zero, this work was recently showcased by Samantha Song and Scott Hanselman at the Microsoft Build 2026 conference.

“Securing AI in the enterprise at pace requires an integrated, full-stack approach,” says Ragini Singh, a partner group engineering manager in Microsoft Digital. “By validating these capabilities together at enterprise scale, we’re generating the real-world learning to strengthen Microsoft’s products and give customers a trusted blueprint for secure AI adoption. Looking ahead, our vision is to make this integrated security foundation the standard for every enterprise, so organizations can scale autonomous agents with speed, confidence, and trust.”

The Secure AI Agents initiative was the result of an all-hands-on-deck project behind the scenes here at Microsoft: Months of testing, coordination, validation, and refinement that transformed an emerging concept into a governable enterprise pattern.

“All of these teams came together and not only enabled the environment in our IT tenant but also ensured that all the right policies were deployed to make it so that these agents can run safely. That meant not bypassing any of the security parameters that we’ve already deployed.”

Shyam Sunder Gogi, technical program manager, Microsoft Digital

Shyam Sunder Gogi, a technical program manager in Microsoft Digital, served as the organizer of the effort, which initially began with a two-week sprint to get ready for Microsoft Build 2026. The project eventually involved more than 70 stakeholders representing different product and business groups.

“All of these teams came together and not only enabled the environment in our IT tenant but also ensured that all the right policies were deployed to make it so that these agents can run safely,” Gogi says. “That meant not bypassing any of the security parameters that we’ve already deployed.”

Testing in real-world scenarios

Rather than validating solutions in an isolated sandbox, we ran a pilot inside Microsoft Digital, standing up a dedicated Windows 365 Cloud PC environment for roughly 100 internal users.

Pilot participants put the solution through its paces across common developer scenarios. These included:

• Windows OS images with Copilot CLI and OpenClaw pre-installed
• Windows 365 Cloud PC provisioning
• Entra Agent IDs to distinguish human and agent identities
• Defender runtime protection and Purview data loss prevention policies
• Intune device and agent configuration policies
• Microsoft Entra Global Secure Access (GSA) network security controls at runtime

Our goal was to pressure-test a full stack of security guardrails, and to do it the way an actual customer would: with real users and real workloads.

“Customer Zero only works if you’re willing to be the first to hit the rough edges,” says Tom McCleery, a principal group cloud network engineering manager on the Microsoft Infrastructure, Network and Tenant (MINT) team here in Microsoft Digital. “We took the agent scenarios into a live tenant with real users, found the issues product teams couldn’t have surfaced in a lab, and fed every one of them back to the team to address before this ever reached broader enterprise readiness.”

The decision to use Windows 365 Cloud PCs in the pilot proved important.

“We want these device agents to run on our employees’ primary machines, and when necessary, to run on a secondary machine that can be easily isolated and reset if needed,” says Dave Rodriguez, a principal product manager for the Endpoint Experience (EE) team in Microsoft Digital. “That’s why we chose to go with Windows 365 Cloud PCs as a corporate-bound, non-primary environment, where we could have our end users work with those machines as they would with any other device.”

The approach created a safe environment for experimentation while giving teams realistic deployment conditions.

“With Windows 365, there’s no real impact to the devices that we’re using,” says Harshitha Digumarthi, a senior product manager on the EE team. “I really love how we leveraged Windows 365 for piloting this, iterating each time there was a change.”

Digumarthi and her team helped to establish and validate all administrative controls and policies, confirm that they function as expected, and execute a phased rollout strategy—starting with pilots and expanding incrementally while proactively monitoring for risks and user impact.

Identity, data, and network controls under load

Much of the runtime security work fell to our Microsoft Digital team, which validated how the controls behaved once agents were actually operating. A foundational aspect was telling humans and agents apart, so that rules and governance can be more clearly defined based on who is overseeing a process.

“Entra Agent IDs let us give an agent its own distinct identity, instead of having it borrow a person’s,” says Joshua Green, principal software engineering manager on the Microsoft Digital team. “Once you can cleanly separate human and agent identities, everything downstream—access decisions, auditing, runtime protection—gets dramatically more trustworthy.”

On the protection and data-governance side, MINT exercised Defender and Purview against agent activity.

“We validated Defender runtime protection and Purview data loss prevention against live agent behavior,” says Diego Baccino, a principal software engineering manager on the MINT team. “It’s one thing to write a DLP policy; it’s another to confirm that it actually catches what an autonomous agent might try to move. That testing is exactly the kind of value that Customer Zero adds.”

Network-layer controls rounded out the stack.

“Entra Agent IDs let us give an agent its own distinct identity, instead of having it borrow a person’s. Once you can cleanly separate human and agent identities, everything downstream—access decisions, auditing, runtime protection—gets dramatically more trustworthy.”

Pradeep Kunjunny, principal PM manager, Microsoft Digital

This was the advantage of using Global Secure Access, showing that the effort worked with traffic loads generated by a real agent.

“Entra Agent IDs let us give an agent its own distinct identity, instead of having it borrow a person’s,” says Pradeep Kunjunny, a principal PM manager in Microsoft Digital. “Once you can cleanly separate human and agent identities, everything downstream—access decisions, auditing, runtime protection—gets dramatically more trustworthy.”

From pilot to platform

Across the initiative, Microsoft Digital drove coordination and execution across multiple organizations.

“Our close collaboration with Microsoft Digital demonstrates the power of validating security capabilities for AI agents in one of the world’s largest and most complex enterprise environments. The insights we gain from real users and workloads help us strengthen our products and give customers greater confidence as they adopt AI agents securely.”

Aakarsh Nair, partner director of engineering, Microsoft Security

Pilot onboarding, validation of Intune-managed control rollouts, and rapid issue-triage loops improved decision confidence before it was demonstrated at Build, and it shaped the product to prepare it for broader enterprise use.

“Our close collaboration with Microsoft Digital demonstrates the power of validating security capabilities for AI agents in one of the world’s largest and most complex enterprise environments,” says Aakarsh Nair, a partner director of engineering in Microsoft Security. “The insights we gain from real users and workloads help us strengthen our products and give customers greater confidence as they adopt AI agents securely.”

The result is a validated blueprint—endpoint, identity, data, and network controls working together—that our customers can now look to as they bring AI agents to their own tenants.

“The winners in enterprise AI won’t just be the teams with the best model experience. They’ll be the teams that make AI operationally trustworthy within the enterprise.”

Tunde Makinde, senior service engineer, Microsoft Digital

It was an initiative that was all about proving that secure AI agent scenarios could work in a real enterprise environment, not just in a demo or a lab setup.

“The winners in enterprise AI won’t just be the teams with the best model experience,” says Tunde Makinde, a senior service engineer for tenant integration and management engineering in Microsoft Digital. “They’ll be the teams that make AI operationally trustworthy within the enterprise.”

Our Secure AI Agents initiative is reinforcing a lesson we’ve learned repeatedly as Customer Zero for the company: Successfully deploying AI in the enterprise isn’t just about delivering innovative capabilities, it’s about ensuring that identity, endpoint, network, runtime, and data protections work together as a cohesive system, enabling employees to use new technologies with confidence while maintaining the governance and security standards that organizations expect.

“By validating these capabilities together at enterprise scale, we’re generating real-world learning to strengthen our products and give customers a trusted blueprint for secure AI adoption.”

Ragini Singh, partner group engineering manager, Microsoft Digital

By bringing together teams from across Microsoft and validating these capabilities in a live enterprise environment, we were able to test how AI agents perform under real-world conditions. The result was more than a successful Build demonstration. It was a practical blueprint that informs how we build our products and provides valuable guidance for companies preparing to adopt agents at scale.

“Securing AI in the enterprise at pace requires an integrated, full-stack approach,” Singh says. “That’s why our team in Microsoft Digital brought together Microsoft Agent 365, Windows 365, Intune, Entra Agent ID and Global Secure Access, Defender, and Purview to secure our agents.”

Working in concert across endpoint, identity, network, runtime, and data, our partnership helped establish the layered security model needed to secure our AI agents.

“By validating these capabilities together at enterprise scale, we’re generating real-world learning to strengthen our products and give customers a trusted blueprint for secure AI adoption,” she says.

Looking ahead, our vision is to make this integrated security foundation the standard for every enterprise, so organizations can scale autonomous agents with speed, confidence, and trust.

The post Securing AI agents in the enterprise: Learnings from our journey at Microsoft appeared first on Inside Track.

At Microsoft, we’ve been working to close that gap with a new kind of agent: Copilot Cowork. Rather than a strict instruction-and-response approach, Cowork has taken the leap to multi-step action across an employee’s Microsoft 365 environment.

From task assistants to true digital coworkers

For years, the promise of AI at work has centered on task assistance, like creating drafts quickly, delivering better summaries, and providing faster answers. But knowledge retrieval isn’t partnership.

“Cowork really sheds light on the art of the possible when it comes to agents, without the need for employees to build them for themselves. It has ignited people’s imagination around what agents can do.”

Stephan Kerametlian, senior director, Microsoft Digital

In Microsoft Digital, the company’s IT organization, we’ve been pursuing deeper impact for AI. We want agents to work alongside our employees, take on multi-step tasks, and produce real outputs, all while keeping the humans that direct their work in the loop.

Copilot Cowork represents that shift. It behaves less like a chatbot and more like a capable teammate that plans, executes, and checks in as it goes along.

“Cowork really sheds light on the art of the possible when it comes to agents, without the need for employees to build them for themselves,” says Stephan Kerametlian, a senior director in Microsoft Digital. “It has ignited people’s imagination around what agents can do, and it helps them understand that they don’t need to be a developer to get high-quality outputs very quickly.”

Microsoft 365 Copilot agents

Copilot Cowork is just one of the agents available through Copilot. Each is most effective in a specific set of scenarios.

Using Copilot Cowork to move from conversation to action

We created Copilot Cowork for workflows that span multiple steps, people, and applications. While traditional chat experiences answer questions or generate content, Cowork can interpret a request, create a plan, gather relevant context, and carry work forward over time.

“Work IQ packages relevance, ranking, and context into something AI can actually act on. It understands who you work with, what you’re working on, and which content matters in a given situation, helping the AI identify the right material and infer the steps needed to deliver an outcome.”

Swapna Malekar, principal product manager, Microsoft Digital

This agent can develop documents, coordinate meetings, generate research, create web applications, and manage ongoing tasks. It also provides visibility into its progress and requests approval before taking sensitive actions.

Work IQ, Microsoft’s intelligence layer for enterprise context, is the foundation of these capabilities. It helps Cowork understand relevant files, meetings, chats, collaborators, and organizational signals so that the agent can identify the right information for the task at hand. Cowork then uses that context to determine the steps needed to deliver the requested outcome.

“Work IQ packages relevance, ranking, and context into something AI can actually act on,” says Swapna Malekar, a principal product manager in Microsoft Digital. “It understands who you work with, what you’re working on, and which content matters in a given situation, helping the AI identify the right material and infer the steps needed to deliver an outcome.”

Our employees are already using Cowork to tackle work that would otherwise require multiple prompts and applications. With a single request, they can create presentations, establish Teams chats for collaboration, schedule recurring follow-up activities, begin building strategy documents, and more.

Cowork can also pull together emails, chats, documents, and news sources into a briefing or transform existing content into an interactive web experience. Throughout the process, employees can refine the work, answer clarifying questions, and approve actions before Cowork moves forward.

Helping employees embrace this watershed moment

During early adoption efforts for Cowork, we learned that successful usage depends as much on behavior change as on technology. Employees who approached Cowork simply as a better chatbot often saw incremental gains.

“Cowork introduces a new way of thinking about work, where you hand off a complex task, close your computer, and the work continues on your behalf. For adoption specialists, the goal is to help employees recognize where that capability fits naturally into their day.”

Petra Glattbach, senior business program manager, Microsoft Digital

People who learned to think in terms of outcomes uncovered far more value. Instead of asking for a draft, they delegated a process. Instead of requesting a summary, they assigned a research task with a defined deliverable.

Agent Launchpad is an instructional program we’ve designed to develop our employees’ agentic AI skills. This effort, alongside other readiness resources, is encouraging people to identify recurring workflows, experiment with multi-step requests, and refine their collaboration techniques over time.

“Cowork introduces a new way of thinking about work, where you hand off a complex task, close your computer, and the work continues on your behalf,” says Petra Glattbach, a senior business program manager in Microsoft Digital. “For adoption specialists, the goal is to help employees recognize where that capability fits naturally into their day.”

This process isn’t about replacing existing ways of working overnight. It’s helping employees recognize where a digital coworker can reduce manual effort, accelerate execution, and create more time for higher-value work.

Boosting our role as Customer Zero with Cowork

Based on our experience as Customer Zero, the most effective Cowork users start with a real business problem and then explore how an agent can help solve it to drive core business outcomes.

For Jody Ryan, principal cloud solution architect for Microsoft 365 Copilot AI Business Solutions, Cowork quickly became part of her daily workflow. She’s used it to build interactive HTML experiences, create adoption sites, and rapidly prototype customer-facing concepts during live conversations.

“Cowork has become a powerful partner in my day-to-day work, helping me turn information into action so I can be more present, proactive, and impactful with my customers.”

Jody Ryan, principal cloud solution architect, Microsoft 365 Copilot AI Business Solutions

In one scenario, Cowork helped her transform a customer discussion into a working web prototype that she refined in real time based on live feedback. Going from whiteboarding to prototype to Agent was a natural progression that helped her customers visualize the real impact of Microsoft’s agentic capabilities.

One of the greatest surprises for Ryan was how much she enjoyed the customizable approach to human-in-the-loop approval checkpoints. For example, Cowork will find a time for a meeting, build a deck, and draft the invite email, but then pause for her review before sending it out. It’s all about identifying patterns of work and enabling Cowork to be part of them.

“Cowork has become a powerful partner in my day-to-day work, helping me turn information into action so I can be more present, proactive, and impactful with my customers,” Ryan says.

Employees in all kinds of roles across Microsoft are echoing Ryan’s experience. People are feeling the genuine evolution that agents like Cowork represent.

The skills we’ve learned over the last three years have been leading to this moment. By demonstrating the tangible impacts of AI through adoption initiatives, celebrating wins, and setting employees free to explore and create AI solutions to business challenges, we’ve positioned ourselves to capitalize on this next leap into more advanced AI tools.

One of our most important lessons has been that cultivating an AI-ready workforce throughout our Frontier Transformation journey has built a sense of confidence and capability with AI. Now that true agentic partnership is a possibility, that journey has prepared our people to get the most value out of tools like this.

The next chapter in the agentic workplace

The response to Copilot Cowork within Microsoft has taken us through an inflection point where AI has moved beyond assistance and into genuine execution. Our internal adoption efforts clearly demonstrate that shift.

“When agents can reason over organizational knowledge and then take action on our behalf, they become a powerful force for productivity and a critical aspect of how we lead Frontier Transformation.”

Brian Fielder, vice president, Microsoft Digital

Within three weeks of its internal release, Cowork had 20,000 users. Employees are actively exploring new ways to use the agent, from streamlining meeting preparation and follow-up work to creating content, conducting research, and managing complex projects.

We’ve learned that different teams often have different uses for different agents. But Cowork is emerging as a tool that can accommodate efforts reaching across a wide array of apps, data sources, workflows, and scenarios.

That’s a big shift, and employees are excited. Interest has been so strong that we’re expanding our enablement efforts, including new, Cowork-focused learning experiences within Agent Launchpad.

As we continue to evaluate Cowork, product feedback is helping us improve reliability, strengthen connections to enterprise data and external systems, and refine the quality of outputs. Even though it’s still in the early stages, teams are finding that Cowork can reduce administrative burden and help work move faster.

“Work IQ is helping unlock a new era where AI can understand the context behind our work, not just the content,” says Brian Fielder, vice president of Microsoft Digital. “When agents can reason over organizational knowledge and then take action on our behalf, they become a powerful force for productivity and a critical aspect of how we lead Frontier Transformation.”

For us at Microsoft, Cowork is more than a new agent. It’s providing a glimpse into a new future of work, where digital coworkers help employees focus more of their time on the aspects of their job that matter most.

The post From AI assistant to capable teammate: How Copilot Cowork is changing the way we work at Microsoft appeared first on Inside Track.

Once captured, meeting data becomes a source of organizational knowledge that helps employees catch up on discussions, revisit decisions, track commitments, and surface unresolved issues. AI-powered experiences such as Microsoft 365 Copilot and Work IQ can connect insights across meetings, helping teams understand how conversations, decisions, and workstreams evolve over time and ensuring important information doesn’t get lost.

Although these features have proven to be incredibly useful to our employees and our wider organization, there are also concerns about how retaining Microsoft Teams meeting data and AI insights might affect our security posture, records retention policy, and privacy. Just like any other company, we at Microsoft must balance these different factors accordingly.

At Microsoft Digital, the company’s IT organization, we’re leading cross-disciplinary conversations on this topic to help ensure that we get it right.

The value of Teams meeting data

Within a meeting, the Microsoft 365 Copilot sidebar experience helps our late-joining employees catch up on what they’ve missed, provides intelligent prompts to review unresolved questions, summarizes key themes, and creates notes or action items.

“The value of a meeting should not end when the meeting ends. Transcription and AI help transform conversations into durable knowledge—making decisions easier to find, commitments easier to track, and information easier to access for everyone who needs it.”

Chanda Jensen, senior product manager, Microsoft Digital

The benefits of AI in meetings extend beyond the live meeting experience as well. When meeting content is available after the meeting, AI can transform conversations into accessible organizational knowledge. Transcripts and underlying documentation—including, notes, decisions, and action items—enable participants to revisit discussions, catch up on missed meetings, verify decisions, and accelerate follow-up work.

Our Microsoft Teams meeting data retention efforts focus on three key categories of artifacts: Underlying documentations, transcripts, and the AI-generated artifacts that help power Microsoft 365 Copilot and Work IQ experiences.

Transcription provides the underlying documentation that makes these AI-powered experiences possible. By making transcription and AI capabilities available in meetings while preserving organizer, administrative, compliance, and sensitivity controls, organizations can choose how these capabilities are used while enabling users to benefit from more effective collaboration and knowledge retention.

“The value of a meeting should not end when the meeting ends,” says Chanda Jensen, senior product manager in Microsoft Digital. “Transcription and AI help transform conversations into durable knowledge—making decisions easier to find, commitments easier to track, and information easier to access for everyone who needs it.”

Policy considerations for meeting data retention

The value of these tools is clear, but data-retention obligations also play an important compliance role that organizations like ours need to consider.

“When individuals generate recordings or other artifacts during meetings, we tend to think of them as an individual’s data, but they actually represent the company’s data. We want to empower individuals, but we have to remember the retention and volume impacts of these artifacts on the company can be substantial.”

Rachael Heade, director of records compliance, Microsoft Corporate, External, and Legal Affairs (CELA)

First, producing and retaining this kind of data can be complex if it isn’t governed properly. For us at Microsoft, this data represents day-to-day general business practices that elevate productivity, and factors such as security and privacy must be considered when managing it. Second, data-rich artifacts like video recordings require a lot of space, quickly eating up cloud storage budgets.

“When individuals generate recordings or other artifacts during meetings, we tend to think of them as an individual’s data, but they actually represent the company’s data,” says Rachael Heade, director of records compliance in Microsoft’s legal division. “We want to empower individuals, but we have to remember the retention and volume impacts of these artifacts on the company can be substantial.”

In light of these potential impacts, some organizations simply opt out of enabling Microsoft Teams meeting recordings.

Asking the right questions to assemble the proper guardrails

Leaders in Microsoft Digital and Corporate, External, and Legal Affairs (CELA), our legal division, are working to balance the benefits of Microsoft Teams meeting data retention with our compliance obligations, aiming to provide empowering experiences for our employees while also keeping company data safe.

“Organizations are always concerned about centralized control over the retention and deletion of data artifacts,” Heade says. “You have excited employees who want to use this technology, so how do you set them up so they can use it confidently?”

Like many policy conversations, getting this right starts with the governance team in Microsoft Digital and our internal partners asking employees from across the company who are responsible for data governance the right questions:

• When should a meeting be recorded and when shouldn’t it?
• What kind of data gets stored?
• Who can initiate recording, and who can access it after the meeting?
• How long should we retain meeting data?
• Where does the data live while it’s retained?
• How can we control data capture and retention?
• What does this mean for eDiscovery management?

These questions help us think about the proper data-retention guardrails. Our IT perspective is only one part of the puzzle, so we’re actively consulting with CELA, corporate security, privacy, the Microsoft Teams product group, the company’s data custodians, and our business customers throughout this process.

“As an organization, this is about thinking through your tenant position and getting it to a reasonable state.”

David Johnson, tenant and compliance architect, Microsoft Digital

Our conversations have brought up distinctions that any organization should consider as they build policy around Microsoft Teams meeting retention:

• The length of time a meeting’s data remains fresh, relevant, or useful
• Consideration of the difference between AI-generated archival content versus the full meeting transcript
• The different risks inherent with recordings compared to transcriptions
• Establishing default policies while allowing limited variability and flexibility when employees require it

“As an organization, this is about thinking through your tenant position and getting it to a reasonable state,” says David Johnson, tenant and compliance architect in Microsoft Digital.

From sharing perspectives to crafting policy

Our policies around Microsoft Teams meeting data retention continue to evolve, but we’ve already implemented some highly effective practices, policies, and controls. Every organization’s situation is unique, so it’s important that you speak to your legal professionals to craft your own policies. But our work should give you an idea of what’s possible through the out-of-the-box features within Microsoft Teams.

The policies we’ve put in place represent a mix of technical defaults, meeting options, and empowering employees to make informed decisions about usefulness and privacy. They also build on the foundations of our work with sensitivity labeling, which helps secure data across our tenant.

Here are some of the practices we follow and features we use:

• Transcript attribution opt-out gives employees agency and reassures them that we honor their privacy.
• Recommending that employees “tell and confirm” before recording empowers and supports our people to speak up when they don’t believe the meeting should be recorded or don’t feel comfortable with this choice. Employees in the meeting can also stop the recording, if needed, or determine if automatic recording was set up but is not appropriate.
• Visual indicators that a meeting is being recorded and that transcription has started, allowing users to request that a meeting stop being recorded or to leave the call.
• User education, through an internal recording smart-use statement document, helps employees understand the implications of recording, when not to record, and when not to speak in a recorded call.
• We do not use compliance recording. While compliance recording could enforce full consent collection, unmuting themselves, we decided that opt-outs and user notices provided sufficient agency to our employees.
• We offer meeting labels that limit who can record, meaning only the organizer or co-organizer can initiate recordings for meetings labeled “highly confidential.”
• Meeting labels are informed by content shared within the meeting. If content is shared in the meeting that has a higher label than the meeting itself, the organizer is prompted to re-label it.
• Meeting labels are inherited and applied to all meeting artifacts, recordings, transcripts, and notes. This means that meeting knowledge remains protected, and any AI consumption of that recording will automatically inform the consumer of the sensitivity and required protections.
• Only meeting organizers can download meeting recordings, keeping the meeting data contained and restricting sharing.
• The default OneDrive and SharePoint meeting expiration is set to 90 days to ensure we minimize the risk of data leakage or cloud-storage bloat.
• The default Meeting AI Archive is set to an 18-month retention policy. That allows questions and decisions from the meeting to be leveraged by the team for post-meeting insights but ensures that data is not kept forever.  
• Deletion is applied in a consistent manner under the business general categories of our retention schedule. The schedule supports our designation of the Teams artifacts as productivity tools and resources, but not as official company records. This stance controls data volume, reduces review and production burden, and ultimately reduces risk (including security and privacy factors).

Balancing productivity with sensible data governance

At Microsoft, we apply different retention periods to different types of meeting data, based on their purpose and business value. Full meeting recordings and transcripts are governed by a default 90-day expiration policy, helping reduce privacy, security, and storage risks while ensuring employees can still benefit from recordings in the near term.

“The bottom line is that we rely on our employees to be good stewards of the company. Because we’ve got a good governance model in place for Teams and solid overall hygiene for our tenant, we’re well set up to deal with the evolution of the product and make these decisions.”

David Johnson, tenant and compliance architect, Microsoft Digital

Separately, AI-generated meeting knowledge—such as questions, decisions, and other insights extracted from meetings and used to support discovery and knowledge-sharing—can be retained for up to 18 months, allowing teams to benefit from the value of those insights long after the original transcript has expired.

These policies are designed to balance employee productivity with responsible data governance, ensuring that important information remains available when useful but is not retained indefinitely. They reflect the three core tenets we use to inform our governance efforts: empower, trust, and verify.

“The bottom line is that we rely on our employees to be good stewards of the company,” Johnson says. “Because we’ve got a good governance model in place for Teams and solid overall hygiene for our tenant, we’re well set up to deal with the evolution of the product and make these decisions.”

The net outcome of all of this work is that our organization is more confident in our approach to meeting knowledge, resulting in more meetings being recorded or transcribed and generating more valuable post-meeting artifacts.

We can’t specifically recommend that an organization follow our blueprint entirely, but asking questions similar to the ones we’ve outlined here can help you build a strong Teams data-governance foundation. With a firm grasp of the technology and close collaboration with key stakeholders, you can guide your own policy decisions and unlock more value for your employees.

The post Empowering employees after the call: Enabling and securing Microsoft Teams meeting data retention at Microsoft appeared first on Inside Track.

“To bring us together, we definitely got creative after COVID, and even more so when AI came in. We did a lot of fun team games where we would share and represent our region.”

Petra Glattbach, senior business program manager, Microsoft Digital

What began as a creative experiment to reconnect people in the post-pandemic era quickly became a way for globally distributed employees to reconnect with one another after years apart.

The comic book—the brainchild of Petra Glattbach, a senior business program manager based in Australia—grew out of a broader effort to rebuild personal connections across a region spanning Australia, New Zealand, Japan, China, Singapore, and India.

“To bring us together, we definitely got creative after COVID, and even more so when AI came in,” Glattbach says. “We did a lot of fun team games where we would share and represent our region.”

For many years before the COVID-19 pandemic, these teams had maintained a strong sense of community through regular regional gatherings and global conferences, creating a rhythm of in-person collaboration and camaraderie that they deeply valued.

Pre-pandemic, companies around the world hosted in-person events to keep their networks strong and make new connections. Remote work was an outlier, and most businesses operated on a face-to-face basis. COVID and its accompanying restrictions transformed professional interactions around the world, platforming remote work as the norm.

Microsoft was no exception.

Maintaining connection in a distributed world

Microsoft 365 Copilot was being rolled out during this time, and employees were encouraged to incorporate it into their daily workflows. Like many people across different businesses and industries, having our employees incorporate AI into daily work was equal parts exciting and anxiety-inducing for the Asia teams.

The employees in the Asia region thus found themselves trying to solve two challenging problems: How do you foster real workplace community when in-person meetups are infrequent and difficult? And how do you build your confidence with AI tools in the workplace? Maybe, they thought, there was a way to tackle both challenges at the same time.

The Asia teams initially focused on the community-building side of the quandary. They started small, organizing gatherings like themed quiz games, cultural trivia, and other online events designed to bridge the gap between the virtual and real worlds.

In the spirit of our company’s AI rollout, the Asia region also began experimenting with different forms of AI. They initially used it for more organizational tasks, such as event brainstorming and planning. However, once they began experimenting, it occurred to them that more creative, community-driven projects might be the perfect opportunity to encourage people to learn about AI while having fun and building closer relationships between geographically separated teams.

“What it really did for us was normalize the idea of using AI. It allowed us to move culturally from trying to use this new technology to do our jobs to the idea that it can help us in many different ways.”

Alan Stone, senior director of business programs, Microsoft Digital

The advent of Hello There

The teams started with some small-scale creative team games and experiences, such as making songs and short videos that represented their different cultures. For instance, every six months one of our teams would run an event they called “Hello There.” They’d randomly assign three people from different parts of the world to find a half-hour in their week for a conversation, and suggest a topic for the group to chat about.

At a time when the company was no longer regularly flying employees together in person, the initiative helped recreate the spontaneous personal connections that had once happened naturally during regional events and conferences. They then asked the participants to use AI to generate an image that told the story of the things they spoke about, which they posted in a dedicated Viva Engage channel.

The generated images were fun for everyone to see and comment on, and they helped employees learn more about one another. The experience gave globally distributed employees an opportunity to connect not just as coworkers, but as people, despite the distance between them.

“What it really did for us was normalize the idea of using AI,” says Alan Stone, senior director of business programs in Microsoft Digital. “It allowed us to move culturally from trying to use this new technology to do our jobs to the idea that it can help us in many different ways.”

Using AI for a creative group project

After a few regional cycles of iterative experimentation with these creative capabilities, our teams in Asia decided to use AI for a more ambitious group project—something that would require teams from multiple countries to work together toward the same goal. They wanted to create an item that represented the Asia region, celebrated their unique cultural attributes, and boosted the AI literacy of their employees.

That’s where Glattbach’s comic book idea came in. Inspired by the popularity of manga and graphic storytelling across Asia, she had the idea of showcasing both the technological capabilities and powerful collaborative nature of AI by creating a comic book using AI-driven tools.

She proposed the project in a meeting with Stephen Kerametlian, a senior director of business program management in Microsoft Digital. Knowing his love of James Bond films, his team created a comic book of Kerametlian as a Bond character, flying to visit employees in countries around the region. Kerametlian says receiving a personalized, AI-generated comic book from team members he hadn’t physically seen in years was a defining and personally touching moment.

“They even managed to get my wife and two-year-old son into the book,” Kerametlian says. “We’re an international team. Through that comic book, I was able to travel the world and ‘see’ my entire team, which I haven’t seen in person since before the pandemic. We had so much fun with it.”

Glattbach’s idea was then translated into a wider-reaching effort across the different teams based in Asia. Groups in different countries would each create a chapter of the shared narrative. Two regional leaders, Alan Stone, and Wai Leong Chan, were main characters in the story, traveling across Asia in a hot air balloon and touching down in each country to visit the team and learn about their culture.

The project enabled team members to learn to use AI while completing a fun and creative effort together. Regional leaders created a purposefully safe, low-stakes environment to encourage experimentation and play.

“It was an opportunity for people have a little bit of fun, knowing they’re experimenting and learning in a safe environment,” says Jane Davis, a director of business programs in Microsoft Digital who’s based in Australia. “I love how easy it was for everybody to get involved and experiment with AI as a shared community.”

The teams prioritized joy and creativity over the final output, encouraging people to work together and fostering an environment where they could learn from one another. Each team shared prompts, compared results, and built on others’ ideas to come up with their country’s chapter of the comic book.

“We unlocked a ton of usage and value from Copilot through something as simple as a comic book. It turned out to be very powerful, and inspired people to use the technology for their own work as well.”

Stephan Kerametlian, senior director of business program management, Microsoft Digital

This larger comic book project was a runaway success. It became a companywide example of successful peer-to-peer learning, cross-cultural exchange, and creative collaboration.

“We were able to rally communities around it; we didn’t have to beg people to join the work sessions, because they were super-popular,” Kerametlian says. “We unlocked a ton of usage and value from Copilot through something as simple as a comic book. It turned out to be very powerful, and inspired people to use the technology for their own work as well.”

Participation grew rapidly; employees were genuinely excited to get involved.

“AI is strengthening our day-to-day productivity,” says Shunsuke Kubota, a senior field IT manager in Microsoft Digital, who is based in Japan. “We’re generating more connections, resulting in more sharing of ideas. I think it’s kicked off an incredible creativity and innovation loop.”

And that’s a win for everyone at Microsoft.

“People think that AI is replacing human connections. We purposely chose to think about creative ways that our community can use AI as an enabler of human connection across geographies in Asia, especially since Microsoft Digital is a global team.”

Jane Davis, director of business programs, Microsoft Digital

Future collaborations with AI

The success of this AI collaboration project has inspired teams across Microsoft to adopt a similar approach to community building, including groups in Europe and the Middle East.

“People think that AI is replacing human connections. We purposely chose to think about creative ways that our community can use AI as an enabler of human connection across geographies in Asia, especially since Microsoft Digital is a global team,” Davis says. “We’re asking, how do you come together as a community? How do you do it in a virtual way, in a way that enables social and experimental learning?”

As these efforts illustrate, when used intentionally and purposefully, AI technology is a powerful tool for amplifying and enhancing human ties in a widely distributed world. As the technology continues to advance, we believe it offers an opportunity to rethink how people connect, learn, and grow together.

“AI is something that’s driving community,” Davis says. “People are discovering these new, innovative opportunities to connect, and AI is their creative partner in making it happen.”

The post How Microsoft Digital Asia used AI to create human connection across continents appeared first on Inside Track.

• Introduction: Building readiness for Agent 365 at Microsoft
• Chapter 1: Establishing a foundation of practice for agent administrators
• Chapter 2: Building a registry of agents to manage them at scale
• Chapter 3: Visualizing agents to support oversight and action
• Chapter 4: Securing agents and aligning Agent 365 with organizational priorities
• Conclusion: Turning visibility into confidence as agents scale

Building readiness for Agent 365 at Microsoft

At Microsoft, we’re on a Frontier Transformation journey to reimagine work and redefine processes through the power of agentic AI.

Microsoft Digital, the company’s IT organization, operates a large and diverse population of agents, built across a broad range of tools and technical capabilities. With Microsoft Agent 365, we now have visibility into more than 500,000 agents.

This distributed control plane has brought agent inventory and governance into one place, giving us a clear view of agent categories, metadata, usage, and ownership information. Agent 365 has also improved our ability to track agent lifecycle and bring in new risk insights.

There’s more work to be done, but it’s already enabling enterprise-scale agent management at Microsoft. Agent 365 provides helpful information about our agent ecosystem, including the top platforms used to create them and the agents our employees use most. It presents this info in helpful, all-up views like the dashboards below.

We’re increasingly connecting agents to business-critical data, involving them in vital workflows, and using them to drive concrete business outcomes. This shift to agentic workflows has inevitably led to questions about operational readiness:

• How do we further accelerate AI-powered innovation without losing visibility, trust, and control?
• How do we create useful, powerful agents while governing them safely?

Agent 365 is becoming an essential vehicle for answering these questions as we enhance agent oversight and control for everyone involved in Frontier Transformation, from AI administrators to security professionals to business decision makers.

Read and prepare

Becoming a Frontier Firm: A guide for deploying AI agents based on our experience at Microsoft.

Agent 365: A response to the challenges of agentic governance

At Microsoft, we share many of the concerns of our customers about properly governing and managing the wide array of agents we build and surface across different platforms. We take a “self-service with guardrails” approach to our productivity estate, which means we give employees the ability to create new workspaces across their Microsoft 365 applications, while we secure assets by default and expand access based on employee needs.

The same is true for agent creation. As a result, the number of agents within our organization has grown rapidly.

“Agent 365 is giving us the confidence to let innovation happen everywhere while ensuring we always understand what agents are doing, how they’re evolving, and where IT needs to engage as a trusted partner in the process.”

Brian Fielder, vice president, Microsoft Digital

Individuals and teams can create agents through a variety of platforms, including Microsoft 365 Copilot Agent Builder, Microsoft SharePoint, Microsoft Teams, Microsoft Copilot Studio, Microsoft Azure AI Foundry, and Agents Toolkit Software Development Kit (SDK). Each platform has its own tools, back-end systems, and ways to view inventory, usage, and risk.

As agents began operating across apps and runtime environments, the need for us to break down management and governance siloes became apparent. An effective method for managing this new class of enterprise asset was required.

We wanted one shared view of all agents in our organization, tightly connected to the people responsible for administration, governance, security, and business outcomes. That’s a challenging prospect—something that no organization has done before.

Microsoft created Agent 365 in response to these needs. In Microsoft Digital, we’ve been working alongside the Agent 365 product team to implement this suite of tools within our production tenant. We’re putting these core capabilities into practice, providing a unified way to observe, manage, govern, and secure agents as they scale across our organization.

“Agent 365 is giving us the confidence to let innovation happen everywhere while ensuring we always understand what agents are doing, how they’re evolving, and where IT needs to engage as a trusted partner in the process,” says Brian Fielder, vice president of Microsoft Digital.

This guide shares what we’ve learned so far:

• How we’re using Agent 365 in Microsoft Digital
• Where we’re supplementing it with additional practices
• Lessons learned that can help you use Agent 365 more effectively, whatever your scale or AI maturity level

From product vision to production

As Customer Zero for Agent 365, it’s important that we’re candid about our journey. Much of the product’s value comes from how we’re incorporating it into our current processes, alongside existing tools.

Scale is also relevant. For smaller or simpler tenants, readiness comes faster. At an organization like Microsoft, with hundreds of thousands of agents, there are times when manual oversight isn’t enough.

We’re actively involved in co-developing the product, uncovering opportunities for capabilities like automation and programmatic solutions to support administration and governance at scale. As part of this process, we’ve partnered closely with the product team to provide continuous feedback and share learnings from our hands-on experiences.

“This has been a strong partnership—daily standups, tracking real issues, and embracing the feedback needed to make the product better. Microsoft Digital plays a critical role as our Customer Zero while operating at a scale like no one else.”

Ray Smith, corporate vice president, Agent 365 product group

Today, we’re using core Agent 365 capabilities while actively sharing feedback with the product group in the following areas:

• Centralizing an accurate inventory of all agents running in the tenant across Microsoft and third‑party platforms to provide a genuinely unified registry across all agent platforms.
• Extending existing enterprise controls by integrating with Microsoft Entra for agent identity, Microsoft Purview for data security and compliance, Microsoft Defender for threat protection, and the Microsoft 365 admin center for operations—all enhanced for improved agent control and management.
• Assisting processes to streamline the lifecycle for agents, including new lifecycle metadata like draft vs. published status, ownership tracking, and usage analysis.
• Surfacing actionable insights and risk signals related to agent behavior, access, data usage, and runtime activity, helping IT prioritize attention and response.
• Supporting enterprise scale through automation and APIs to help manage large, diverse agent deployments without relying on manual management.

While full lifecycle capabilities for certain agent platforms, risk signals, and enterprise-scale automation evolve, we continue to partner with the product group to close gaps while existing processes support current operations.

“This has been a strong partnership—daily standups, tracking real issues, and embracing the feedback needed to make the product better,” says Ray Smith, corporate vice president for the Agent 365 product group. “Microsoft Digital plays a critical role as our Customer Zero while operating at a scale like no one else.”

Read and prepare

Governing AI agents at scale: Lessons from our journey at Microsoft 

Chapter 1: Establishing a foundation of practice for agent administrators

A new opportunity to break down silos between roles

As we began scaling agents inside Microsoft, we discovered that the future of agent management would need to evolve from our current ways of working. We wanted a world where we could create and use agents broadly while keeping administration manageable and consistent.

Getting there required new patterns of practice for IT, especially for administrators operating across different focuses. Agent 365 unifies observability between enterprise roles, acting in concert with the broader Microsoft suite of administration, security, identity, and governance tools.

Here is a summary of the needs of different personas involved in the agent-building and management processes, grouped by office and broken down by role:

Office of the CIO

Developers and makers

Build, test, and deploy intelligent agents at scale

Products: Agent Builder, Copilot Studio, Microsoft Foundry

IT administrators

Control, govern, and monitor agents across the organization

Product: Microsoft 365 Admin Center

Agent users and business decisions makers

Get work done faster with AI-powered assistance

Products: Copilot and Teams

Office of the CISO

SecOps

Detect threats and secure agent activity in real time

Product: Microsoft Defender

Data and compliance

Protect data and enforce compliance policies

Product: Microsoft Purview

Identity manager

Manage identities and access agents and users

Product: Microsoft Entra

We’ve found that our most effective AI administrators come from existing Microsoft 365 backgrounds, because they already have deep expertise with mature tools and processes. Whether they’re generalists or specialists, your administrators will already be positioned to manage agents at scale and use their skills and experience with the tools and insights that Agent 365 delivers.

Shifting from siloed administration to coordinated responsibility

Up until this point at Microsoft, managing agents has been the responsibility of the platform administrators who control agent creation tools. SharePoint administrators manage SharePoint agents, Power Platform administrators manage Copilot Studio agents, and so on. Meanwhile, identity, security, and compliance teams handle their respective layers using Microsoft Entra, Microsoft Defender, and Microsoft Purview—often independently.

We know that this model has the potential to break down as agents become more powerful and more interconnected, and as new agent types begin to run autonomously with their own identities.

“With agents in action across multiple spaces, managing them is a special challenge. It was clear that we needed a silo-buster to govern this new ecosystem effectively.”

Jonathan Clare, principal service engineering manager, Microsoft Digital

Within Microsoft Digital, we’re using Agent 365 to differentiate agent management from platform-specific administration without replacing existing expertise. Instead of creating a single, centralized agent manager that encroaches into each platform’s territory, Agent 365 gives us one shared view across platforms, so administrators can coordinate their work with the same data and context.

With this new, single pane of glass, we’re building shared responsibility and clear handoffs where they make the most sense. For example, our AI administrators manage the full lifecycle of Microsoft 365 Copilot Agent Builder agents. But when Copilot Studio is involved, they collaborate with Power Platform administrators to strategically manage those agents in their specific environment.

Agent 365 provides the connective tissue by providing details and common metadata as we move between platforms and administrators.

“With agents in action across multiple spaces, managing them is a special challenge,” says Jonathan Clare, principal service engineering manager in Microsoft Digital. “It was clear that we needed a silo-buster to govern this new ecosystem effectively.”

Evolving agent management from existing roles

One key insight we’ve uncovered from this work is that agent administration doesn’t require a new IT skill set. It builds on the same foundational experience we already use to manage products like Power Platform, SharePoint, Exchange, and Entra, or other identity-based systems.

“We’re still iterating on the seams between administrators with different responsibilities. Agent 365 is providing the space for AI and identity administrators to work closely alongside their colleagues in security, compliance, and governance.”

David Johnson, principal PM architect, Microsoft Digital

The core skills of maintaining accurate inventory, ensuring visibility and access, managing lifecycle, and mitigating risk are already mature and deeply established in our organization. Agent 365 now gives us the broad insight we need to oversee all agents in one place.

From there, we can lean into our well-developed expertise and mature processes with newly enhanced tools, shared metadata, logging, and controls. This coordination gives each team in sequence a sense of clarity and partnership, rather than feeding effort up and down a chain of approval.

“We’re still iterating on the seams between administrators with different responsibilities,” says David Johnson, a principal PM architect in Microsoft Digital. “Agent 365 is providing the space for AI and identity administrators to work closely alongside their colleagues in security, compliance, and governance.”

As we progress, we’re developing a three‑part administrative model facilitated by the oversight that Agent 365 provides.

AI administrators, the primary users for Agent 365

• Oversee complete agent inventory and usage at the tenant level
• Manage the agent lifecycle with the platform coordination to cover a broad range of agent types
• Provide the connective tissue between security, governance, identity, and platform administrators

Agent Identity administrators, new with Agent ID

• Manage agent identities after provisioning and throughout the agent’s lifecycle
• Manage lifecycle events tied to users, access changes, and deprovisioning
• Build identity backed policies for agent workload management and risk mitigation

Security, compliance, and governance teams

• Define the guardrails that apply to agents and agent blueprints, portable specifications for agents’ identities, capabilities, constraints, policies, data access, and lifecycles
• Approve the kinds of data, tools, and permissions agents can request
• Set agent evaluation and risk assessment criteria along with risk-aligned approval workflows
• Align publishing and runtime decisions with risk tolerance and security policy

Agent 365 facilitates this model by providing comprehensive agent coverage. This acts as a shared coordination layer, bringing different administrator, security, identity, and governance roles into a unified space.

Agent 365 in practice: Agent publishing and workflows

We didn’t create Agent 365 to handle every IT workflow. Many approval, vetting, and escalation processes are specific to an organization’s risk posture and operating model. At Microsoft, we’re currently handling much of that logic using an existing risk assessment and publishing workflow while evaluating how Agent 365 capabilities can simplify those steps. An example of the type of risk we look for is when an agent could read sensitive data and write it to destinations with broad access, like external sites or apps.

There are several areas of risk we use Agent 365 to assess:

Agent 365 itself assists us with additional risk awareness:  

Real-time

• Observability across agents
• Surfacing signals from identity, security, and governance systems
• Supporting the ability to act when risks or issues surface

Proactive

• More intelligent risk insight during the agent permissioning and approval processes
• Consistent agent publishing into the environment
• Forthcoming capability: the ability to integrate with our existing agent review and publishing process that spans multiple teams, including administration, governance, and security

As you consider ways to collaborate across your own administrator teams, our silo-busting approach can act as a helpful guide.

Chapter 2: Building a registry of agents to manage them at scale

A centralized source of truth for AI agents across the enterprise

As agents have proliferated across Microsoft, visibility has proven essential for robust governance. Without a clear understanding of all the agents that exist in our environment, including their origin and how people use them, it’s very difficult to make informed decisions or respond confidently when risks emerge. Establishing a thorough registry of agents and their key information is a critical step in governing the ecosystem.

Agent 365 provides that oversight.

Why an agent registry matters

An agent registry establishes the foundation for oversight, control, and compliance. As an organization introduces more agents, the environment can quickly become fragmented and difficult to track.

A comprehensive registry provides a single, authoritative inventory that makes every agent visible, tracks ownership, and captures key metadata. With that baseline, organizations can consistently govern, secure, and manage their agents with confidence.

“Managing agents begins with having a complete inventory with rich information, like their name, lifecycle status, type, ID, owner, where we created them, and where we’re using them. Once you have that level of clarity, everything else—security, compliance, lifecycle management—becomes much easier to manage.”

Mike Powers, AI administrator, Microsoft Digital

At the scale of a company like Microsoft, even small gaps in visibility can quickly become operational hurdles or compliance liabilities. Without that foundation, an organization faces substantial risks:

• Ownerless agents remain active after employees leave the company.
• Shadow or unsanctioned agents are difficult to detect.
• Oversight is unreliable with respect to agent growth, usage, and impact.

“Managing agents begins with having a complete inventory with rich information, like their name, lifecycle status, type, ID, owner, where we created them, and where we’re using them,” says Mike Powers, an AI administrator in Microsoft Digital. “Once you have that level of clarity, everything else—security, compliance, lifecycle management—becomes much easier to manage.”

Agent 365 registry capabilities

Within Agent 365, the registry acts as a central inventory enriched with metadata. We use that metadata to meet compliance, security, and management standards, including:

• Agent ownership and associated teams
• Creation platforms and publishing surfaces
• Lifecycle states like “draft” or “published”
• User scope: who can discover and use the agent

In Microsoft Digital, we’re providing real-world feedback to the Agent 365 product group to enable even more types of metadata, like differentiation between system vs. user-created agents, and new agent types like AI teammates. We’re also using metadata surfaced from the platform where the agent was created, for example, the Power Platform environment ID for Copilot Studio agents and the Azure AI Foundry subscription details for Foundry agents.

Agent 365 amalgamates all of this metadata into a single view.

This structure matters because agents vary widely. Some are short‑lived experiments. Some belong to only one employee, while others are broadly shared. Treating them all the same doesn’t make sense.

From an administrative perspective, the registry gives us:

A summary view of total agent count

Insights around growth and adoption

Identification of agents without owners

Analytics on platforms used to make agents and their usage trends

Search, sort, and filtering with customizable columns to get detailed views across agent types

Robust inventory export capabilities to support collaboration with security, compliance, and business stakeholders

Establishing our Agent 365 registry

The agent registry is an out-of-the-box feature for Agent 365, so there’s nothing to deploy or configure. As Customer Zero, we’ve focused much of our early work on validating the registry for accuracy and completeness.

Agent 365 automatically ingests agent metadata from supported platforms. This technology is still new, so we’ve partnered with product teams across SharePoint, Power Platform, Azure AI Foundry, and other builder experiences to reconcile counts, ensure accuracy, and request additional relevant metadata.

For first-party tools, Agent 365 creates registry entries automatically. Third-party agents can also benefit from automatic registration if their creators use the Agent 365 SDK during development.

For pro‑code scenarios, Entra Agent ID is key. Registering an agent through Entra assigns it a formal identity, which lays the groundwork for consistent identity and lifecycle management and conditional access policies.

Acting on the registry

The registry is a living system. Ownership changes, while lifecycle states and usage signals update automatically.

That means the registry supports critical processes for administrators that include:

• Passing audits for elements like tracking agent ownership
• Presenting the tenant’s agent footprint and usage to business decision makers
• Scoping agents to specific users, or excluding users based on regional or regulatory requirements
• Highlighting high‑impact agents based on usage and runtime

A single view has been one of the most valuable outcomes for us, enabling informed operational decisions and peer-to-peer collaboration.

Looking ahead

The registry is also the prerequisite for future experiences, including broader agent discovery and publishing. Moving forward, it will provide the context we need to guide reuse, review, publishing, and eventual retirement to support intentional agent lifecycle practices over time. As a result, it will be easier to combat sprawl and ownerless agents.

In Microsoft Digital, our early Agent 365 efforts have focused on validating our agent registry to lay the foundation for comprehensive observability. It may be helpful for you to mirror this approach.

Chapter 3: Visualizing agents to support oversight and action

Observability: Scaling beyond dashboards

At the scale of an organization like Microsoft, dashboards alone aren’t enough. We already have hundreds of thousands of agents in use across the company. At that scale, it would be impossible to review these agents individually. We rely on well-established governance in the form of guardrails, established software development lifecycle procedures, and risk-based app and agent management policies that trigger reviews when we detect risk.

Agent 365 helps us operationalize oversight using automation and rules engines, programmatic access via APIs and scripting, and bulk actions based on attributes like permissions, connectors, and usage patterns.

A simple user interface is essential for visibility, assessment, and decision‑making. Programmatic access is essential for execution. Effective agent administration requires both.

The lesson is that administering agents during Frontier Transformation requires a new approach that breaks out of traditional roles and inter-team hierarchies. By incorporating our experience into your own planning, you can use Agent 365 more effectively.

Why visualization matters

As agents spread across Microsoft, we learned that inventory alone isn’t enough. Knowing an agent exists is helpful, but understanding how people use it, how it connects to data and other agents to complete workflows, and where risks or concentration points emerge is what makes effective governance possible at scale.

In a Frontier Firm where almost anyone can create agents, observability is a core pillar of management. Like many organizations, we built agents first and only later confronted the challenge of seeing what existed. Agent 365 will help other organizations reverse that order by surfacing agent behavior continuously from the start.

“Just this first layer of visualizing our agent ecosystem in one central place is a big step toward flowing them into our business processes and demonstrating ROI more effectively.”

Amy Ceurvorst, director of business programs, Microsoft Digital

Visualization is helping us address questions we couldn’t answer before:

• Where is agent growth accelerating?
• Which agents are widely used?
• Where do risk hot spots occur across connectors, data sources, and permissions?
• What demands attention now, and what can wait?

“We’re uncovering so many new use cases for agents,” says Amy Ceurvorst, a director of business programs in Microsoft Digital. “Just this first layer of visualizing our agent ecosystem in one central place is a big step toward flowing them into our business processes and demonstrating ROI more effectively.”

The agent landscape changes quickly, and with Agent 365, we can look at usage at the individual agent level to track shifts over time. For example, Cowork (Frontier) is one of our newest agents, but in just a few weeks it became our most widely used.

In a recent review of Cowork adoption, Agent 365 allowed us to quickly analyze names, session activity, and locations for Cowork’s 58,000 active users in just a few minutes.

This is also where Agent 365 complements rather than replaces Viva Insights:

• Viva Insights combines Agent 365 data with our organization’s people data to provide enhanced insights into agent usage across the organization.
• Agent 365 provides oversight for the full agent estate: registry, publishing, ownership, lifecycle, and governance.

Both are important, but they serve different personas. Where Viva helps clarify usage for adoption leaders and change managers, Agent 365 helps determine what action IT should take next.

The Viva Insights Agent Dashboard extends the data in Agent 365 by translating agent inventory and telemetry into executive‑ready insights on adoption, usage patterns, and trends across the organization. By combining agent activity with organizational context, it helps leaders understand where people are using agents, how adoption is evolving over time, and where opportunities or risks may exist.

Together, Agent 365 and Viva Insights provide a governed, end‑to‑end view that supports informed decisions about scaling and governing agents to drive business impact.

From insight to action

One of our biggest lessons as Customer Zero is that visualization only matters if it leads to action. In Agent 365, insights increasingly surface as prioritized scenarios, such as risky, ownerless, or unused agents. We can then pair those insights with paths to response—for example, meeting compliance expectations by re-assigning or retiring ownerless agents.

“Agent 365 is saving us time by helping us analyze the kinds of issues that are common in agent management and bringing those to our administrators’ attention. It acts as a command center that surfaces those issues programmatically, so we’re able to prioritize the actions we need to take.”

Nate Zimmer, senior product manager, Microsoft Digital

Visualization in Agent 365 is about prioritization. For us, some of the most valuable scenarios include:

Ownerless agents

Visualization helps us distinguish unused agents we can retire vs. heavily used agents that need reassignment.

Duplication and naming patterns

Many agents share similar names. Visual signals help differentiate them based on usage and connections.

Trending agents

Rapid adoption can signal value or risk, necessitating a closer look for administrators.

Strategic adoption signals

Identify high-impact agents to scale as business priorities.

The goal is to focus attention where it counts.

“Agent 365 is saving us time by helping us analyze the kinds of issues that are common in agent management and bringing those to our administrators’ attention,” says Nate Zimmer, a senior product manager in Microsoft Digital. “It acts as a command center that surfaces those issues programmatically, so we’re able to prioritize the actions we need to take.”

Continuously clearing the fog

Observability is never finished. New agent types and creation tools continue to emerge. Agent 365 helps us embrace that reality by connecting signals across Microsoft 365, identity, security, and compliance for a continuously updated view of an evolving agent landscape.

We believe the Agent Map is going to be a differentiator in agent visualization, and we’re closely working with the product team to inform new, robust capabilities that will help us drill down to find hot spots, details on agent connectors, tools, and knowledge sources more easily.

For example, we use the large surface area provided by the Agent Map to search and filter for exactly what we want and then dive deeper into details.

Observability has been crucial for helping us guide agent usage at Microsoft. As you conduct Frontier Transformation at your organization, consider ways that observability has led to better oversight for our team, and incorporate them into your AI administrators’ discipline.

Chapter 4: Securing agents and aligning Agent 365 with organizational priorities

Melding agent oversight, identity, security, and governance

In Microsoft Digital, we’ve learned that securing agentic AI isn’t about inventing an entirely new security model. Instead, the focus should be on extending the identity, data, and threat protections we already trust, while also making risk visible in one place. Agent 365 plays a critical role by surfacing agent‑related security signals in a single view so that IT teams can see what matters quickly, even when remediation happens elsewhere.

The agentic security challenge

Up to this point, understanding agent risk has meant pulling information from multiple tools and manually stitching together context. Identity management lives in one place, data protection in another, and threat insights somewhere else. That makes it harder for IT administrators to spot patterns and gain insight.

“A lack of visibility creates real security risk, exposed data access, unmonitored behaviors, and unmanaged identities. Agent 365 helps us regain control by building a comprehensive inventory and risk profile of agents.”

Prathiba Enjeti, principal security manager, CISO

Other factors compound the challenge:

• People and teams are creating agents quickly, accelerating the need for manual reviews.
• Agents can operate across apps, data sources, action types, data, and data destinations, and they carry the potential for other agents to expand the attack surface, complicating oversight and control.
• Risk emerges at multiple stages, both during agent development (design time) and during execution (run time).

“This lack of visibility creates real security risk, exposed data access, unmonitored behaviors, and unmanaged identities,” says Prathiba Enjeti, a principal security manager for the Microsoft CISO organization. “Agent 365 helps us regain control by building a comprehensive inventory and risk profile of agents.”

Theoretically, existing agent governance policies and practices should mitigate these risks, but there are always exceptions. It’s easy to miss early warning signals, and teams may only detect issues after they have an impact.

Agent 365 helps us identify and remediate those issues.

Agent 365 as a security visibility layer

As we bring Agent 365 into our operational workflows, it connects with Microsoft Purview, Microsoft Entra, and Microsoft Defender, surfacing relevant agent‑specific risk insights in a cohesive experience. That reduces fragmentation and supports more informed, coordinated decisions.

You can see how Agent 365 capabilities apply to different members of the agent administration and management ecosystem.

Rather than replacing those tools, Agent 365 ingests identity signals from Entra, data signals from Purview, and runtime behavior from Defender. In practice, we think about agent security in two main categories:

• Build‑time risk: These signals surface when people create or configure agents. Examples include overly broad permissions, insecure configurations, or missing responsible AI safeguards. Seeing these early helps reduce downstream risk and rework.
• Run-time risk: As agents operate, they can expose data unexpectedly, become susceptible to vulnerabilities like prompt injection, or lose protection as data moves across systems. Run-time visibility becomes even more important as agents begin working together.

Agent 365 doesn’t eliminate these risks, but it does have the capacity to make them more visible, traceable, and easier to prioritize and mitigate. Follow‑up actions still happen in Entra, Purview, and Defender, but now those administrators benefit from improved oversight and coordination.

What we’ve learned so far

Internally, broader visibility has helped us uncover issues we had difficulty tracking before, like ownerless agents spanning multiple platforms or unexpected data handling behaviors. While more broadly available agent oversight might seem intimidating because it widens scrutiny, we’ve found that additional data and insights have accelerated alignment and improved decision making.

With Agent 365, we now have better conversations through shared context and metadata. As a result, IT, security, and business teams can discuss adoption trends and mitigate risk using the same information instead of chasing it across tools.

Conclusion: Turning visibility into confidence as agents scale

As we reflect on the early days of Agent 365, visibility is the foundation for everything that follows. As Customer Zero, our priority has been to understand the full extent of agents across our environment.

The real value will come when we can drill down further. How are people using agents? What risk patterns repeat? What building and usage trends emerge as Frontier Transformation progresses?

“When we started, I thought the hardest part would be the technology, but it turned out to be building the weekly rhythm where IT, security, identity, product, and business unit teams could look at the same picture, make decisions from the same facts, and act together. My advice to every IT leader is to establish that governance rhythm before your agent count grows faster than your ability to manage it.”

Garima Tiwari, principal product manager, Agent 365 Customer Zero, Microsoft Digital

As we mature in our use of Agent 365, it will give us greater ability to move from simple metrics like volumes of agents to more meaningful measures of impact. It will also help us see trends in our environment by segmenting low-use experimental agents from business‑critical digital workers so we can move beyond isolated usage to scaled adoption.

It’s important to be clear about where we are on this journey. Our operating model for Agent 365 isn’t complete. Much of our current focus is still on seeing clearly by surfacing trends, comparisons, and emerging patterns we couldn’t identify before.

“When we started, I thought the hardest part would be the technology, but it turned out to be building the weekly rhythm where IT, security, identity, product, and business unit teams could look at the same picture, make decisions from the same facts, and act together,” says Garima Tiwari, a principal product manager for Agent 365 Customer Zero in Microsoft Digital. “My advice to every IT leader is to establish that governance rhythm before your agent count grows faster than your ability to manage it.”

That visibility is already paying dividends by revealing duplication, ownerless agents, and cross‑platform risks that would otherwise remain uncovered. Over time, those insights will increasingly feed automation, lifecycle workflows, and deeper integrations across identity, security, and governance.

Looking ahead, our next steps are about translating this oversight into action at scale. That will include preparing for broader agent discovery, refining lifecycle management, and enabling new personas, such as managers who are responsible for selecting, creating, and overseeing digital workers. It will also encapsulate learning as new agent types, tools, and usage patterns emerge. Change is constant in a Frontier Firm environment; readiness is something you build continuously, not something you check off once.

“Agent 365 represents a new operating model for AI at scale. By bringing visibility, governance, and security together, it helps organizations move beyond experimentation and toward a future where agents are trusted and embedded in everyday work, all without slowing innovation.”

Stephan Kerametlian, senior director, Microsoft Digital

Our overall message in this guide is straightforward: You don’t need to have every answer on day one. What matters most is establishing the conditions for safe evolution as agents scale. Think about clear administration practices, a reliable registry, effective observability, and security signals you can trust. Here at Microsoft, Agent 365 has become an important part of that foundation.

“Agent 365 represents a new operating model for AI at scale,” says Stephan Kerametlian, a senior director in Microsoft Digital. “By bringing visibility, governance, and security together, it helps organizations move beyond experimentation and toward a future where agents are trusted and embedded in everyday work, all without slowing innovation.”

We’ll continue sharing what we learn as Customer Zero. As your organization moves through its own Frontier Firm transformation, we hope these lessons help you build the confidence to innovate quickly, supported by comprehensive insights, thoughtful governance, and security that scales with your ambition.

The post Implementing Agent 365: How we’re governing and managing AI agents at Microsoft appeared first on Inside Track.

During in-person briefings, I often remind our customers that generative AI systems like Copilot and Copilot Cowork are still quite new, and that they upend decades of human-computer interaction patterns. For over 40 years, we’ve trained knowledge workers to memorize patterns in the GUI to support their productivity.

With Copilot and agents, you can type nearly anything as a prompt and instantly get a detailed response, or create an agent to autonomously complete a business process on your behalf. It’s a very new way of working, and it’s no surprise that many knowledge workers—and even engineers themselves—are still adapting.

More than anything, our customers are looking for one key insight: They want to know how we measure and define the business value of Copilot and agents at Microsoft, especially through the lens of increased productivity and cost savings. After all, AI systems represent a significant investment for our customers, and leaders want to have confidence that they’ll yield sufficient ROI to justify the cost, knowing that the payback period isn’t immediate.

Six steps to proving value

So, how do you prove the value of AI-powered tools in the enterprise? There are six main steps that I share with customers that I believe are critical, based on the empirical evidence we’ve gathered in Microsoft Digital.

Our big mistake? We didn’t instrument all the arduous business processes that slow us down and impact effectiveness before our deployment, so that we could more easily prove the value of AI across our enterprise after deployment.

Here are further details on the six measurement steps to keep in mind:

1. Before you deploy, talk to people who are doing the frontline work in your organization. These should be hands-on managers or influential individual contributors (ICs), not executives or mid-level managers who are abstracted away from the work (although these individuals can be useful in helping identify the right SMEs to talk to).

For each role, identify three to five everyday pain points that could benefit from Copilot or agents. These could be operational, business, or technological processes, all of which can be improved by thoughtful applications of AI. This phase provides a great chance to use Six Sigma skills or other continuous improvement (CI) methodologies to evaluate your processes and identify waste, then consider how a combination of CI and AI could make them more efficient.

2. Once identified, instrument and measure each of the processes to get baseline data on the average time it takes to complete them. Then the hard work begins—you need to think deeply about how AI could make those processes better. This could be through defining a series of step-like prompts that take away the toil. It could be through further AI-powered automation to make some of the steps go away. It could be applications of AI that validate outputs to ensure that rework is minimized. More and more, it could be an agent that completes the work on the employee’s behalf.

If you lack detailed telemetry for the process in question, you have two options. The first is preferable: Build end-to-end telemetry, so you have observability throughout the process in question. While this is always the best way to reliably measure productivity gains at scale, there is also a second option: Identify a cross-section of ICs and measure them the old-fashioned way as they complete the process in question—with a stopwatch. Take the average of several people to get a better sense of how much time is typically needed to complete the task. Then consider how AI could make that process more efficient, while also improving the resulting business outcomes.

3. Concurrent with your investigation into ways that AI can improve productivity and reduce toil, you need to invest in enterprise AI skilling, ideally by role. The fact is, engineers are going to use AI differently than operations staff, who are going to use it differently than sales and marketing pros. Yes, there are common skills for each, but the best training is tailored to the role, grounded in the experience of using Copilot or agents to address challenges or opportunities that commonly arise in their day-to-day work.

It’s also important to ensure your employees understand the strengths and weaknesses of current AI models. No model is perfect, and understanding where reasoning errors can occur will help them avoid accepting AI-generated output that may be inaccurate. Human discernment and observability is a critical step in validating AI outputs.

A best practice is to gate access to generative AI; require employees to engage in both general and role-specific training prior to their provisioning to ensure they immediately unlock value in their work.

4. After you’ve trained your employees and given them the skills to succeed with AI and agents in the enterprise, begin your deployment in earnest. For tools like Copilot, we recommend deploying in cohorts by role, focusing on employees who will immediately see the greatest benefit. At Microsoft, we started with our sales team and then gradually deployed to the remaining employee population over the course of several months. This enabled us to scale up our skilling programs, governance strategies, and support function in anticipation of increased volumes. For agents, deploy to support the biggest pain points first, then monitor and observe agent performance before scaling to additional business or operational processes.
5. Post-deployment, give it a few weeks, then return to those same processes you identified in the first step and measure the average time savings. If time savings aren’t as significant as you hoped with either Copilot or agents, go back to process evaluation and employee skilling and continue to refine your approach.

This whole process is intentionally iterative—you’re not likely to get it exactly right on the first attempt. But if a process that used to take 30 minutes can now be completed in 10, you’ve obviously made a big difference that will save a lot of time when extrapolated across an entire fiscal year. Even better if that end-to-end process is now being completed by an autonomous agent, with a human reviewing and validating the output.

After you’ve succeeded in one domain, identify different role leaders and find another batch of processes that could be improved with a combination of continuous improvement and AI. There’s really no limit to the scope of ways you can positively impact your employees if you maintain a sharp focus on continually improving their experience with Copilot and agents.

6. The final step: Reclaim the value from those productivity savings and apply it to new business challenges or opportunities. Let’s be clear—most enterprises aren’t investing in AI solely to give their people back time. They’re trying to do more with less, enabling their workers to be more productive and generate more value for the company.

In this crucial final phase, you need to be thoughtful and deliberate about how you’ll use the time saved with AI. If your people are saving two hours per week on average, how will they use that time? There are innumerable ways you could redeploy that time to address new business challenges or opportunities. The key is to be intentional in maximizing value, so it aligns with your team and your company’s goals and ambitions. Then report those savings to the appropriate members of the leadership team, to help them understand how their AI investments are paying off.

There you have it—you now possess the tools to quantify the value of your generative AI investments in the enterprise, using them to recapture value and drive more business impact thanks to the power of Copilot and agents.

Learn more

Dive deeper into this topic with a framework that our organization put together to measure the impact Copilot is having at Microsoft.

Boosting AI adoption with structured change management

As you can see from the graphic above, it’s important to surround the AI value measurement process with structured change management and ongoing employee skilling. In Microsoft Digital, we’ve learned that even the most useful or intuitive technologies won’t see widespread adoption without a deliberate and sustained change management effort that’s localized to meet the disparate needs of a global organization.

In our case, that meant cultivating and supporting a community of Copilot Champions who have become the backbone of our global change management strategy. This community is now 10,000+ strong and even has its own Viva Engage forum, where our Copilot enthusiasts answer employee questions and share useful prompts.

Unfortunately, just doing AI skilling once won’t be enough. The pace of change with Copilot and agents is simply too great to do one training effort and then move on. Building an AI-forward culture takes time, and ongoing opportunities to learn and improve skills are one of the best ways to help your employees build the AI habit.

The Microsoft 365 Admin Center has an “AI adoption score” that can help you see—by cohort—how successfully people are building that habit. Having employees who use Copilot three times per week in any way is enough to build an AI-focused workforce, enabling your company to unlock the value of AI in the enterprise.

The promise of generative AI is significant, and there’s no doubt that you’ll see qualitative benefits in your organization even if you don’t instrument every process. But in an era of tight IT budgets, having the quantitative data necessary to calculate the ROI of your investments in Copilot and agents will make it far more likely that you’ll get financial support from your executive team to deploy and succeed at scale.

The post Unlocking the value of Microsoft 365 Copilot and agents at Microsoft appeared first on Inside Track.

As these processes grew in scale and complexity, it became clear that improving something highly manual and already operating at massive scale would require a fundamentally different approach.

“With BPO, we’re dealing with high-volume, high-touch processes that are core to how the business runs,” says Jonathan d’Orgee, an AI transformation lead for Microsoft Business Operations.

For many organizations, the idea of overhauling a core business process can feel like a daunting step. At Microsoft we act as our own first customer, which gives us a way to test, refine, and de-risk that transformation in our own operations before bringing those proven patterns to customers. We call this approach Customer Zero.

In this case, that meant rethinking how high-volume operations could run better with AI directly embedded into day-to-day tasks, including building solutions using tools like Microsoft Dynamics 365 and Azure AI.

“We looked at manual steps, broken workflows, and disconnected systems as opportunities for AI transformation.”

Jonathan d’Orgee, AI transformation lead, Microsoft Business Operations

Learn more

Check out more stories about our Customer Zero journey with our products and services here at Microsoft.

Identifying manual inefficiencies

On top of the complexity of handling so many transactions across the globe, Business Operations sometimes experienced periodic surges that could exacerbate inefficiencies. During these surges, the team would see a high volume of complex, time-critical transactions— especially at the end of the month or the quarter—and manual processes were too slow to keep up.

As we reviewed these inefficiencies, we looked for the most impactful use cases—places where we could integrate AI into workflows. To do this, we asked two important questions:

• What types of transactions have the highest volume?
• What parts of the process take the longest time or consume the most resources?

It was a classic case of the 80/20 rule—finding the 20% of the processes that required 80% of the work.

“We looked at manual steps, broken workflows, and disconnected systems as opportunities for AI transformation,” d’Orgee says.

An example might be where we receive an email asking to have a contract updated. In the former process, the email might sit there until a human could review it manually. Then someone would review it, direct it to the right queue, and assign it to the right person.  

“With AI in the workflow, emails and attachments are analyzed right when they arrive, and immediately assigned to the right queue and person,” d’Orgee says.

Taking these kinds of steps dramatically increased efficiency and reduced costs overall.

“With deep knowledge of our Business Operations ecosystem, we targeted high-volume, repeatable workflows across globally distributed operations. These were processes where AI could break traditional location and labor constraints, unlocking scalable automation and measurable business impact.”

Shashidhar Lanka Venkata, partner group engineering manager, Business Commerce Platforms

Configuring an AI toolkit

Once we’d identified the areas that were ripe for transformation, we set about developing an AI-driven solution on top of our existing critical workflow systems.

“With deep knowledge of our Business Operations ecosystem, we targeted high-volume, repeatable workflows across globally distributed operations,” says Shashidhar Lanka Venkata, a partner group engineering manager in the Business Commerce Platforms team. “These were processes where AI could break traditional location and labor constraints, unlocking scalable automation and measurable business impact.”

The BPO AI Toolkit is our AI operating system for business process operations. Its job is to help us with decision making. Built on Microsoft Dynamics 365 and Azure AI, it brings process mining, Microsoft 365 Copilot, Windows 365, and the Azure Marketplace together into AI-native workflows that can be reused by different vendors.

The toolkit is built on a handful of capabilities that work together:

Agentic memory turns tribal knowledge into structured operational intelligence that agents can access on demand.

Prebuilt agents provide enterprise-ready capabilities that teams can reuse instead of rebuilding workflows.

An agentic UI reduces context-switching time, helping operators focus on decisions and exceptions.

Digital Twins measures real end-to-end process performance and continuous improvement.

Agent Desktop provides secure access anywhere.

“It’s just part and parcel of working with AI, which is much different than working with more traditional ways of automating,” says d’Orgee.

He explains that because the AI is configurable, our teams are able to move faster. “The lead time is a lot shorter, and we’re able to make changes a lot more quickly.”

At the core of everything during this effort was the drive to constantly assess “the human buy-in:” How are people using this technology in a way that solves real problems at a global scale?

Keeping humans in the loop and measuring AI transformation

Integrating AI into existing workflows and processes isn’t just about the technology—it also should entail a cultural shift within an organization.

We wanted to ensure that our operations team was adopting the AI tools in the right way. That meant understanding which processes must still be human-led, such as areas where the handling of exceptions requires more discernment.

Rather than removing humans from the process, the team redefined the human role. AI now handles tasks such as data validation, case creation, and compliance checks, while our team members focus on judgment, exceptions, and continuous improvement.

“It’s really exciting for us, because operations has always been about trying to be efficient. With AI, it’s allowed for breakthroughs that we haven’t been able to achieve before.”

Jonathan d’Orgee, AI transformation lead, Microsoft Business Operations

That balance helped the team scale automation without losing the oversight and expertise needed to maintain quality.

The impact of this Frontier model has been significant. So far, we’ve been able to transform roughly a quarter of our BPO processes with AI. This has led to an 80% improvement in process quality and a 33% reduction in cost per transaction, d’Orgee says.  

More than 75% of the cases our teams work on are processed utilizing the AI toolkit. These gains are measured with Digital Twins, a process-mining model that monitors each workflow live, allowing teams to continuously track and improve. Building on this momentum, the team has plans to transform 80% of the BPO process with AI by fiscal year 2028.

D’Orgee urges organizations that want to apply our Customer Zero learnings to their own workflows to look for high-volume, high-effort, highly manual work. This will lead you to the best opportunities for automating your processes at scale and deliver the most benefit.

From finance to sales operations, teams across Microsoft have turned to the BPO AI toolkit to prove how reusable AI capabilities can drive enterprise-wide transformation.

“It’s really exciting for us, because operations has always been about trying to be efficient,” d’Orgee says. “With AI, it’s allowed for breakthroughs that we haven’t been able to achieve before. I’ve just been thrilled to come to work on that front.”

The post Streamlining business operations at Microsoft with an AI toolkit appeared first on Inside Track.

This makes accurate content essential. If our content is stale, even the best agent or search engines will return wrong answers, which is frustrating to everyone.

“Knowledge management today is about making sure people can find the right answers the moment they need them. When knowledge stays current, employees get unblocked faster and productivity improves, making the overall support experience far more efficient.”

Silvina Olkies, senior director, Service Management, Microsoft Digital

That means the knowledge bases that get tapped for answers need be accurate, and for that to happen, they need to be updated frequently and without delay.

Internally here at Microsoft, that’s where our team got involved.

We’re Microsoft Digital, the company’s IT organization, and our team saw an opportunity to use AI to dynamically and proactively update our knowledge management systems.

Our first step was—in partnership with our Global Help Desk—to strengthen our self-serve help and reduce the number of steps users need to take to find the right answers. It’s a process that many of our own customers can apply to their knowledge management transformation.

“Knowledge management today is about making sure people can find the right answers the moment they need them,” says Silvina Olkies, a senior director of Service Management in Microsoft Digital. “When knowledge stays current, employees get unblocked faster and productivity improves, making the overall support experience far more efficient.”

The challenge: Fragmented knowledge, manual reviews

For our Global Help Desk, the challenge wasn’t just the volume of content, but also its condition. Support knowledge is spread across thousands of self-service articles, agent-facing systems, and SharePoint sites, all constantly evolving.

“If the knowledge isn’t accurate and current, the experience breaks down immediately. Bad content leads to bad answers.”

Kevin Verdeck, senior IT service manager, Microsoft Digital

In today’s fast-changing AI-powered world, it doesn’t take long for knowledge to become incomplete, out of date, or redundant. This shows up in the inaccurate answers employees might receive.

In an environment increasingly powered by search and AI, weak knowledge equals weak results.

“If the knowledge isn’t accurate and current, the experience breaks down immediately,” says Kevin Verdeck, a senior IT service manager in Microsoft Digital. “Bad content leads to bad answers.”

At our Global Help Desk, keeping that content current required a manual review process.

Our teams analyzed usage data, depended on support agents to report missing or outdated content, and worked through recurring review cycles that relied on subject-matter experts to help confirm whether articles were still accurate. This took significant time and coordination, and even then, some issues were identified only after employees had already hit a dead end.

The result was a system that was reactive and hard to scale.

When employees couldn’t find answers, issues were pushed to advanced support. Poor knowledge quality created poor outcomes, while those responsible for fixing it were struggling with maintaining it.

“One five-member team was reviewing 1,900 self-service KB articles and 1,700 agent-facing KB articles every six months, and that didn’t even include the many SharePoint sites,” Verdeck says. “It was basically their full-time job doing regular reviews.”

Turning raw data into knowledge

Our team in Microsoft Digital set out to build AI for Knowledge Management, a centralized system that could scale across multiple repositories, cut the manual work of keeping content current, reduce reliance on busy content owners, and prepare knowledge for people and AI to use.

“When you have a large volume of data, it’s a silent gold mine. The sheer brilliance lies in taking that data, making it sing, and letting it tell you exactly where the treasure is.”

Ankit Guddewala, software engineer II, Microsoft Digital

An AI pipeline solution made sense because we wanted to fix the issue at scale.

The real opportunity was to turn every resolved support ticket into a signal that looked at what the employee was asking for (nature of the issue or request), whether the answer was already documented, and how the AI and human agents handled it. So, we began with our large amounts of support ticketing data and systems as the starting point.

“When you have a large volume of data, it’s a silent gold mine,” says Ankit Guddewala, a software engineer in Microsoft Digital. “The sheer brilliance lies in taking that data, making it sing, and letting it tell you exactly where the treasure is.” 

The stages to complete the work happen as follows:

1. Ingest the raw support data: The team pulls in large volumes of incident data from our ticketing systems.
2. Clean and structure noisy data: Tickets can include conversation notes, incomplete details, inconsistent writing styles, and abandoned issues. We use the AI enrichment layer to turn those details into structured fields, such as reported versus actual problems and remediation steps.
3. Find patterns across incidents: We cluster tickets to help identify recurring issues and avoid cluttering the knowledge base with one-off scenarios.
4. Compare patterns against existing knowledge: We use the system to search current articles and rank how closely the resolution aligns to current content to determine what steps to take next. For example:
   • Below 40 percent: Create knowledge
   • 40 to 80 percent: Update existing knowledge with missing details
   • Above 80 percent: No update needed
5. Notify the appropriate knowledge managers: Subject-matter experts are notified by email so they can validate the change(s) and add more detail if needed.

“Our goal is to free people from the manual work of maintaining content so they can focus on improving its quality. With the right human-in-the-loop balance, AI can do the heavy lifting while people make sure the final knowledge is accurate and useful.”

Namrata Ladda, product manager II, Microsoft Digital

The result is far less time spent reviewing thousands of articles during every review cycle. We keep humans in the loop to validate the output. Their feedback helps tune the AI model, so it improves over time.

“Our goal is to free people from the manual work of maintaining content so they can focus on improving its quality,” says Namrata Ladda, a product manager in Microsoft Digital. “With the right human-in-the-loop balance, AI can do the heavy lifting while people make sure the final knowledge is accurate and useful.”

Impacts and what’s next on the journey

Using our new AI for Knowledge Management platform, our Global Help Desk teams can identify knowledge gaps without waiting for someone to report them. They’re using AI to generate structured article drafts so humans can focus on quality, not search through data.

The Global Help Desk projects the solution will save them an estimated 16,000 hours annually; result in a 10 percent reduction in support tickets; and reduce the number of advanced support escalations. This leaves everyone on the team more time to directly help employees more quickly, when needed.

“Turning our knowledge base from a static thing into a living knowledge base is a big step forward,” Ladda says. 

Other Microsoft teams, including HR, have expressed interest in leveraging the content management platform. Once the product has completed internal testing, AI for Knowledge Management will be released to all company employees and customers.

“This solution moves knowledge management from manual maintenance to an intelligent capability that helps organizations scale and apply what they know more effectively,” Olkies says.

The post AI for Knowledge Management: Keeping support content up-to-date at Microsoft appeared first on Inside Track.

At Microsoft, this shift is grounded in years of investment in our neurodiverse employees. From launching our Microsoft Neurodiversity Hiring Program to building a company-wide focus on neuro-inclusive employee experiences, we bring our lived experience to how we design and build our products.

We set out to better understand where our products could do more for all our users and challenge established accessibility norms.

“When we create space for neurodivergent voices to shape our products, we don’t just improve accessibility, we build better technology for everyone. Accessibility is strongest when it is informed by lived experience, and initiatives like our product feedback sessions make sure those voices directly influence how our products evolve.”

Neil Barnett, chief accessibility officer, Microsoft Accessibility

Neurodiversity Celebration Week became the launch point for a new approach to product feedback and co-design at Microsoft. We used the event to create an ongoing program that brings the perspectives of neurodivergent employees directly into the product development process, especially individuals who experience challenges with executive functioning tasks such as focus, learning, memory, organization, and task completion.

For our first feedback round, 150 neurodivergent employees evaluated 10 products across multiple divisions using real-world scenarios. Rather than focusing only on traditional accessibility checks, they shared insights on how products support everyday thinking, learning, and productivity.

The feedback provided product teams with a deeper understanding of where experiences can be simplified or made more intuitive and easier to navigate. More importantly, it established a repeatable process for gathering perspectives from people across the cognitive spectrum, helping teams build products that work better for more people.

“When we create space for neurodivergent voices to shape our products, we don’t just improve accessibility, we build better technology for everyone,” says Neil Barnett, chief accessibility officer in Microsoft Accessibility. “Accessibility is strongest when it is informed by lived experience, and initiatives like our product feedback sessions make sure those voices directly influence how our products evolve.”

How neurodiversity can help a broader audience

Neurodiversity refers to the idea that neurological differences—including autism, ADHD, and dyslexia—are a natural part of human diversity. It recognizes that there’s no single “right” way for the brain to work.

“Neurodivergent employees bring an innovative way of thinking and have ideas that help make our products better.”

Tarena Shanaberger, senior PM, Microsoft Accessibility

Product groups at Microsoft value feedback from our neurodiverse employees. However, without a consistent way to get it, product groups had to reach out with ad hoc requests through the different inclusion networks at the company.

We saw an opportunity to create both lasting change and more efficiency in this process.

“Neurodivergent employees bring an innovative way of thinking and have ideas that help make our products better,” says Tarena Shanaberger, a senior PM on the Microsoft Accessibility team.

Designing with lived experience

Neurodivergent inclusion is most valuable when it’s built in early in the product lifecycle, preferably before assumptions take hold and shape the product in ways that are harder to change later.

“Lived experience is a genuine form of expertise. You can have telemetry, design reviews, usability metrics, and KPIs that mean things to people who review data, but there are aspects of cognitive load and human behavior that become much more visible when you involve people who experience systems differently.”

Karl Niblock, architect, Engineering and Architecture Group Security

Karl Niblock, an architect in our Engineering and Architecture Group Security team, believes that early intent can result in experiences that delight users by making products “shockingly” easy and a joy to use, even for something as routine as logging into a tool.

Designing systems that help reduce unnecessary cognitive friction means more people can do their best work consistently and with more confidence. Those are the types of human-centered insights that aren’t possible to derive from data alone.

“Lived experience is a genuine form of expertise,” Niblock says. “You can have telemetry, design reviews, usability metrics, and KPIs that mean things to people who review data, but there are aspects of cognitive load and human behavior that become much more visible when you involve people who experience systems differently.”

When people share their experiences, everyone can learn and benefit.

“One of the patterns I often see is dead-end workflows, where a user follows the process exactly as instructed but ends up stuck, with no obvious next step,” Niblock says. “For someone with dyslexia or autism, the challenge is often not the task itself but the repeated effort of decoding similar-looking instructions and trying to determine what went wrong. By designing clear recovery paths, plain-language guidance, and visible next actions, we can dramatically reduce cognitive friction. Those improvements help neurodivergent users, but they also make products easier and less stressful for everyone.”

Building a better path for product feedback

Software engineer Jordan Cowe surfaced an idea to host an annual neurodiversity bug bash to uncover where products pose challenges for neurodivergent users. He explains that neurodivergent employees are often some of the first people to identify points of friction before release and says that the bug bash helped shine attention on these issues and encourage product group action on the feedback.

“If something affects a neurodivergent employee, it likely also affects many people who don’t identify as neurodiverse. Fixing these issues improves the products for everyone, making them easier to use and better at keeping users engaged.”

Jordan Cowe, software engineer II, Copilot Engineering team

This bug bash led to a structured way to bring in testers, distribute feedback results to product groups, and infuse key learnings into our products. Our testers asynchronously went through the instructions, followed the scenarios, recorded their screens, talked through their pain points, and completed a product survey. We used the results to rate product usability, which helped us identify issues and ideas for improving our products.

“If something affects a neurodivergent employee, it likely also affects many people who don’t identify as neurodiverse,” Cowe says. “Fixing these issues improves the products for everyone, making them easier to use and better at keeping users engaged.”

What feedback looks like from the product side

Microsoft AI UX researcher Audrey Aday used feedback from the co-design sessions to explore what makes an AI response feel inclusive, useful, and manageable. Tester feedback showed that the issue wasn’t capability, it was control.

“Instead of treating accessibility as a check-the-box exercise, we’re building relationships with neurodivergent employees and embedding their feedback into the full product development lifecycle. We’re moving away from ‘test this for us once’ to ‘build this with us, continuously.’”

Audrey Aday, UX researcher II, Microsoft AI Design

Testers consistently said Copilot can feel overwhelming when it returns too much information at once. They want more control (not less capability) over how much detail they see. In response, product teams are exploring custom instruction menus, smarter defaults, and adaptive personalization that learns individual preferences without creating extra work.

Product teams across Microsoft see lived experiences as a quality driver, not an edge case. Neurodivergent employees help teams spot usability opportunities that benefit everyone, especially around information overload, pacing, and clarity.

“Instead of treating accessibility as a check-the-box exercise, we’re building relationships with neurodivergent employees and embedding their feedback into the full product development lifecycle,” Aday says. “We’re moving away from ‘test this for us once” to ‘build this with us, continuously’.”

How the Inclusive Tech Lab supports co-design

Support also comes from the Inclusive Tech Lab, where Microsoft design teams work directly with users with disabilities to uncover exclusionary designs and identify new product opportunities.

“The best time to engage with us is when employees are starting something new or revising old designs. The Inclusive Tech Lab helps teams partner with disabled advisors early in the design process.”

Sarah Heinzen, senior designer, Microsoft Design and Research

The lab connects teams with people who bring their lived experience across different scenarios, such as limited mobility, photosensitivity, low vision, and neurodivergence. This co-design approach brings in users early and keeps them involved so their lived experience can shape product decisions from the start.

“The best time to engage with us is when employees are starting something new or revising old designs,” says Sarah Heinzen, a senior designer on the Microsoft Design and Research team. “The Inclusive Tech Lab helps teams partner with disabled advisors early in the design process.”

Building feedback into how we work

While the product teams work on incorporating feedback, we’re hard at work solidifying this continuous feedback loop and adding it to everyday product development at Microsoft.

Partnering with the Inclusive Tech Lab, we’re setting the foundation for making that a reality using:

• One consistent path to feedback. Product teams have a unified way to engage with employees from the disability and neurodiversity communities.
• A participant pool of testers. We’re building a diverse network of volunteers to make sure diverse lived experiences consistently inform product decisions.
• Inclusive ways of working. We’re establishing guidelines to make sure testing scenarios match real user needs.

The expectation is start early, stay engaged, and build with the people you’re designing for.

“Even as a software engineer and someone who’s neurodivergent, accessibility used to feel like something you thought about later,” Cowe says. “Now I pause early and ask: Does this make sense for real users? Is it clear? Is it usable? It’s changed how I build. I’m looking for the small friction points, the things people struggle with, and I’m catching them sooner. It’s made me a better engineer.”

The post Boosting accessibility at Microsoft with help from neurodivergent employees appeared first on Inside Track.

To make this ideal a reality, our team in Microsoft Digital—the company’s IT organization—created an agentic AI assistant that we’re now using to connect new devices to our network. We built the agent into GetConnected, the internal portal that 18,000 of our employees, vendors, and network administrators use each month to register their new devices to our network when they turn them on for the first time.

Our new workflow is simple, fast, and intuitive—and it’s a significant step up from our previous experience.

Creating the GetConnected AI assistant is part of the role we play as the company’s Customer Zero, where we test and use our technology and platforms first and then share our lessons learned with customers. In this case, we’re sharing how we used the Microsoft Agent Framework (MAF) to enhance onsite device management for our employees. GetConnected does not apply to remote device registrations.

“We’ve been looking across our set of services and capabilities to find places where we can do some experimentation leveraging AI. We wanted to be able to test our hypothesis around those AI investments and then be able to double down if it proved correct.”

Jason Thompson, principal PM manager, Microsoft Digital

Improving a highly trafficked internal tool using AI

Our employees use GetConnected to ensure their wired and wireless devices are registered on the network, as well as to extend device expiration dates and check on the status of their devices.

Heavy employee traffic and the repetitive actions users tend to take on GetConnected led us to realize that the tool was the perfect candidate for an agentic transformation. Our goal was to turn what was a five- or six-step process into something that could be completed in just one or two actions.

“We’ve been looking across our set of services and capabilities to find places where we can do some experimentation leveraging AI,” says Jason Thompson, a principal PM manager in Microsoft Digital. “We wanted to be able to test our hypothesis around those AI investments and then be able to double down if it proved correct.”

The team decided to start small, focusing on a couple of the most popular and crucial functionalities within GetConnected.

“We’d seen previous projects that were very ambitious fail because they tried to achieve too much at one time. Based on customer feedback, we noticed that registration is the simplest, most common action that users were having trouble with. So we said, ‘Let’s do that first.’”

Aayush Dave, product manager, Microsoft Digital

It was also important to listen to our employees—our Customer Zero frontline users. They told us which actions in the experience mattered most to them.

“We’d seen previous projects that were very ambitious fail because they tried to achieve too much at one time,” says Aayush Dave, a product manager in Microsoft Digital. “Based on customer feedback, we noticed that registration is the simplest, most common action that users were having trouble with. So we said, ‘Let’s do that first.’”

Next up was deciding how the agent would appear in the portal. The Microsoft 365 Copilot model of a sidebar chat menu worked well for other workflows, so it seemed appropriate to approach the new GetConnected experience in a similar way. This enabled us to create a new experience alongside the existing workflow, so customers could still access the original process (should they need to) and compare the two experiences.

“Other teams might decide to automatically replace the UI with an agent,” says Faris Mango, a principal software engineering manager in Microsoft Digital. “But that’s hard, because now you’re forcing people to use the agent. If it’s not ready to be used at full capacity, they don’t have an alternative to accomplish what they intended. We wanted to avoid that situation.”

Testing out Microsoft Agent Framework (MAF)

To build the agent, we considered two paths.

The first was to directly call the Model Context Protocol using JavaScript, an option that would require significant amounts of coding on our part.

“Some declarative agent systems do all of the things in the background, and you don’t get to turn all the little knobs. MAF gives you the flexibility to make the experience exactly what you want.”

Darron Sullivan, principal software engineer, Microsoft Digital

The second was to use Microsoft Agent Framework (MAF), which proved to be simpler and more customizable for our needs.

“Some declarative agent systems do all of the things in the background, and you don’t get to turn all the little knobs,” says Darron Sullivan, a principal software engineer in Microsoft Digital. “MAF gives you the flexibility to make the experience exactly what you want.”

The tricky part, however, was that MAF was fairly new at the time. In fact, the week that the team started developing the GetConnected agent was the same week that MAF was released in preview internally. As we were building out our agent, the framework was going through its own updates, which threatened to hinder our progress. Even one small change to the framework could break our tool’s entire functionality.

“They were moving really fast, and we were adopting new features and finding new bugs all the time,” Sullivan says. “You had to go through that rapid iteration and development, which is a challenge, but it was also pretty awesome because we’re working on the cutting edge.”

The upside was that we were able to provide valuable feedback to the MAF engineers, which in turn could supercharge the work we were doing on our agent. As a bonus, our partnership drove other teams to pursue similar projects.

“The knowledge sharing across our org was notable and crucial,” Dave says. “Our team was one of the first to start building a solution like this, and we presented in numerous architecture forums to share the components and frameworks we were using, and the teams we were working with. This brought the tide up for all boats in our organization, encouraging other teams to start kicking off similar projects as well.”

Building a seamless, discoverable interface

The agent currently has several key functions, the most prominent of which is to register a device on your behalf.

Previously, employees would have to fill in a long, complicated form that asked for a lot of technical details that they often didn’t know offhand, like type of device or the preferred network.

Instead of just selecting options and approving, the flow is more conversational. The user can start with a suggested prompt like “Help me register a device,” and the agent will ask for the required information (with examples for each field). If the user isn’t sure about something (for example, how to find a MAC address), they can ask follow-up questions, and the agent will pull in FAQ and help content to guide them.

Once all the required details are collected, the agent can complete the registration on the user’s behalf after the user approves it.

Once it has your approval, the chatbot submits the request and replies whether or not it was successful. Users can also ask the agent to show devices that are expired or will soon expire, then prompt the agent to renew those devices if desired.

Seamlessly integrating the agent into GetConnected required upgrading the existing user interface using Fluent.

These updates were needed to support the AI interface integration. Specifically, we introduced a custom header action to launch the AI side panel. Prior to upgrading, doing this would have required using Coherence components outside of their intended patterns.

“We used Fluent AI components to build the AI interface. This helped ensure a consistent Microsoft look and feel across the experience, built-in accessibility for scenarios like screen readers and mobile usage, and components that are designed for conversational and agent-driven interactions.”

Nathan Chambers, software engineer, Microsoft Digital

To stay consistent with the existing app architecture, we upgraded core dependencies like Fluent UI and Coherence to their latest versions. As part of upgrading Coherence across several major versions, it also required us to move the feedback experience to Centro to align with the updated patterns. We then needed to update other parts of the experience like navigation, FAQ, and release notes to match those newer component patterns.

“We used Fluent AI components to build the AI interface,” says Nathan Chambers, a software engineer in Microsoft Digital. “This helped ensure a consistent Microsoft look and feel across the experience, built-in accessibility for scenarios like screen readers and mobile usage, and components that are designed for conversational and agent-driven interactions.”

While making these upgrades, we ran tests to ensure the experience was accessible—for example, for screen reader users or others who might access GetConnected on their phones.

“You can have an amazing, strong piece of software that is well built and focuses on security. But if you don’t have the traffic or people are not using it, it’s worthless.”

Faris Mango, principal software engineering manager, Microsoft Digital

Next, we wanted the agent to be as discoverable as possible. Without people actually navigating to it, there would be no way to show proof of concept. So, we built it so the agent automatically popped open via a side panel when someone loaded GetConnected.

“You can have an amazing, strong piece of software that is well built and focuses on security,” Mango says. “But if you don’t have the traffic or people are not using it, it’s worthless.”

We also wanted to gather early feedback from users. Before releasing it to the entire company, we had internal team members and frequent GetConnected users give the agent a try. Almost immediately, it was clear we had too many approval notices.

“At the beginning, we would have approvals for every single action. For example, if you wanted to see a device in different regions like Puget Sound, Latin America, or Canada, you had to do a separate approval for each region,” Dave says. “This was a very painful experience. So we removed all the approvals and pared it down to a one-click experience.”

Users also had issues with the approval language, which they said was hard to understand and looked like an error message. The next iteration took out much of the technical jargon, making the message more conversational and easier to read.

An agent experience driven by feedback

Our work as Customer Zero is never done. For GetConnected, we’re eager to keep collecting feedback. One major goal is to improve the agent’s performance, making it faster and more responsive.

Our feedback survey is tied directly to a performance dashboard, which tracks metrics like new and returning users, total unique users, conversions, and interactions. Each user submission generates a work item.

“When users leave feedback about something they don’t like, I feed that to the team, and then we sit down and figure out how we can improve that specific part of the experience,” Dave says.

With each update, we’re seeing the payoff of more users and more interactions. The traditional method of registering a device is also seeing a drop-off as more people lean on the agent for assistance.

“Before, you used to have to go into the system and change something about the experience manually. Now, our engineers are going to an AI model and telling it, ‘Hey, you’re doing this part wrong, please improve it.’”

Aayush Dave, product manager, Microsoft Digital

Looking ahead to more use cases

Building an agent has allowed the team to embrace an entirely new type of engineering.

“Before, you used to have to go into the system and change something about the experience manually,” Dave says. “Now, our engineers are going to an AI model and telling it, ‘Hey, you’re doing this part wrong, please improve it.’”

It’s also serving as a reminder to seek progress over perfection.

“AI is changing things so quickly,” Thompson says. “It’s better to do rapid prototyping and roll it out, and start getting the data in terms of how successful the experience is. Then you can let that guide you, in terms of how you iterate going forward.”

Because of the success we’ve had with the GetConnected device registration feature, we’re already exploring other capabilities, including bulk operations to accommodate our facilities managers who need to onboard many devices at once. As AI agents become mainstreamed in many workflows across our organization, we anticipate usage and functionality will continue to grow exponentially.

The post Simplifying device registration at Microsoft with an agentic AI assistant appeared first on Inside Track.

At Microsoft Digital, the company’s IT organization, we’re using a set of employee councils and connected capability groups to guide and accelerate how we deploy and adopt AI across our enterprise. Our goal is to focus our energy on the AI-enabled scenarios that matter most, reducing duplication, strengthening accountability, and making sure our investments create measurable value.

“Our AI decisions and direction must be grounded in business strategy. AI councils provide guidance and enablement for our organization, ensuring that our investments in AI generate tangible benefits to our business. It’s not just developing technology and then looking for a problem to solve with it—we start with the opportunity.”

Don Campbell, principal group technical program manager, Microsoft Digital

That focus matters, because AI success doesn’t come from usage alone. It comes from connecting strategy, enablement, data readiness, responsible AI, continuous improvement, change management, and measurement into one driving force.

That’s how we’re moving from experimentation to repeatable outcomes and from AI enthusiasm to AI accountability.

“Our AI decisions and direction must be grounded in business strategy,” says Don Campbell, principal group technical program manager in Microsoft Digital. “AI councils provide guidance and enablement for our organization, ensuring our investments in AI generate tangible benefits to our business. It’s not just developing technology and then looking for a problem to solve with it—we start with the opportunity.”

Our council-based approach is helping us accelerate our Frontier Firm transformation. The councils work together to set direction for AI adoption at Microsoft Digital, ensuring that our business needs drive solution development that can keep up with the pace of AI change. This work includes building visibility into all our AI solutions, including agents and Model Context Protocol (MCP) servers, while establishing governance and proven practices; developing training and learning pathways; and connecting teams together that are working on similar solutions across the enterprise.

We’re excited for a future where our employees use intelligent agents and human judgment together to work smarter, move faster, and unlock new value for Microsoft and our customers.

Why we use councils to guide internal AI efforts

Effective AI needs both enterprise guidance and business-owned direction. That’s why we’re using councils and connected capability groups as the operating model for our AI deployment.

Each group has a distinct role, and none of them work alone. Together, they help us connect the strategy for AI to the work currently happening across Microsoft Digital.

• Our strategy council sets priorities by aligning AI work to business goals, identifying top scenarios, prioritizing investments, and keeping KPIs and value in focus.
• Our enablement council uses our AI Center of Excellence to turn strategy into action through technical guidance, proven practices, ideation, learning, knowledge sharing, culture, and governance.
• Our data council strengthens the AI foundation via data strategy, governance, access, quality, literacy, and prioritization.
• Our process council drives continuous improvement through operational excellence, problem solving, prioritization, value realization, coaching, and learning.
• Our compliance council applies Responsible AI principles to ensure compliance, inclusiveness, fairness, transparency, and reliability.
• Measurement ties it all together by tracking both business outcomes and engineering artifacts, ensuring we can clearly demonstrate real-time value realization.  

These councils help us see across the business landscape through the lens of AI. They enable us to reduce duplication, scale what works, and make better decisions about where AI can create value. That allows our teams to keep moving fast without letting activity get ahead of accountability.

Aligning AI strategy to business value

Our strategy council helps us decide which AI-enabled scenarios deserve the most attention, which investments align to our business priorities, and how we’ll know whether the work is creating value. It gives leaders a practical way to look across the portfolio and keep our AI work tied to the outcomes we’re accountable for.

“Business strategy defines the what and the why. AI defines the how, enabling execution of the strategy and delivering real value. We should use AI to advance our business strategy, not the other way around.”

Qingsu Wu, principal group product manager, Microsoft Digital

This is important, because broad experimentation is useful early on in your AI journey. It helps teams learn and build momentum. But experimentation has to mature into focus. Without that shift, organizations can end up with too many different agents, agent skills, MCP servers, and other artifacts, without a clear view of what’s actually impacting the business.

We’re using the strategy council to keep that from happening.

“Business strategy needs to lead the AI strategy,” says Qingsu Wu, a principal group product manager in Microsoft Digital and an influential member of the strategy council. “Business strategy defines the what and the why. AI defines the how, enabling execution of the strategy and delivering real value. We need to use AI to advance our business strategy, not the other way around.”

That principle shapes how we work. We use the strategy council to identify our top AI-enabled scenarios, clarify the value we expect to create with each one, and connect that work to a monthly operating rhythm. Product owners still manage delivery and the council keeps the portfolio focused, visible, and aligned.

Tuning strategy into repeatable execution

Our AI Center of Excellence (CoE) is at the heart of our approach to enablement. It helps us translate enterprise AI priorities into practical guidance and execution support for teams building AI-enabled solutions.

“We can see patterns that a single team can’t. We’re translating AI CoE strategy and enterprise priorities into clear execution plans that work in each organization’s context. That allows us to align priorities and make sure our biggest bets are actually landing.”

Ria Khetan, senior program manager, Microsoft Digital

The AI CoE extends the reach of the strategy council. It gives teams what they need to build, govern, reuse, and scale what matters, while the strategy council assists us in deciding where to focus.

That connective role is central to the broader council model. The strategy council identifies the top AI-enabled scenarios. The AI Center of Excellence connects strategy to execution across the organization, operating as a cross-functional coordination layer that sets direction and creates shared accountability.

“We can see patterns that a single team can’t,” says Ria Khetan, a senior program manager in Microsoft Digital, who is a member of the council. “We’re translating AI CoE strategy and enterprise priorities into clear execution plans that work in each organization’s context. That allows us to align priorities and make sure our biggest bets are actually landing.”

The COE helps teams move those scenarios forward with answers to important questions:

• What initiatives are in flight?
• What initiatives bring the most return on investment?
• Where is there potential duplication?
• Where do we need clearer guidance?
• Where do we need stronger governance?

It also helps reduce fragmentation. When teams build in isolation, they can solve the same problem in different ways. They can choose different patterns, interpret standards differently, or create solutions that don’t scale beyond a single context. Enablement gives us a shared way to look across that activity and ask better questions.

“We use the CoE to bring consistency to how AI work gets done,” Campbell says. “It gives us a way to step back and ask whether we’re solving the right problems and whether we’re set up to scale.”

“High-quality, well-governed data is essential to accelerate AI implementation and adoption, and to ultimately unlock its full value. Data quality, accessibility, and governance are imperatives for AI systems to be reliable, scalable, and business-critical. Recognizing this principle is propelling our data strategy.”

Miguel Uribe, principal PM manager, Microsoft Digital

Building AI on trusted data

Our AI scale depends on trusted and reliable data. That makes our data council central to our council-based approach. This council makes sure our teams work with data that’s governed, discoverable, accessible, and ready for AI.

“High-quality, well-governed data is essential to accelerate AI implementation and adoption, and to ultimately unlock its full value,” says Miguel Uribe, a principal PM manager in Microsoft Digital and member of the data council. “Data quality, accessibility, and governance are imperatives for AI systems to be reliable, scalable, and business-critical. Recognizing this principle is propelling our data strategy.”

We’re applying a data mesh mindset to balance domain ownership with enterprise consistency. Teams stay close to the data that they know best. Shared standards for governance, quality, metadata, and compliance provide a framework to make that data useful across Microsoft Digital.

Microsoft Fabric and Microsoft Purview are key to that approach. Microsoft Fabric unifies our siloed data in a shared data mesh. Microsoft Purview enables governance and best practices to ensure that we manage our data responsibly through discovery, classification, protection, and monitoring.

Our goal is AI-ready data that’s available, complete, accurate, and high quality. Our data council also works with the AI Center of Excellence to strengthen data and AI fluency through learning pathways, operational practices, and community programs.

“Our capacity to drive process improvements has been crucial to our AI transformation as a company. We’ve adopted a ‘CI before AI’ approach to ensure that we don’t end up automating inefficient processes.”

David Laves, director of business programs, Microsoft Digital

Improving the process before applying AI

AI works best when it’s applied to the right problem. That’s why continuous improvement is part of our council-based approach. Before teams automate a workflow or build an agent, we want them to understand the process, identify waste, and decide where AI can create measurable value.

“Our capacity to drive process improvements has been crucial to our AI transformation as a company,” says David Laves, director of business programs in Microsoft Digital and a member of the Continuous Improvement Center of Excellence. “We’ve adopted a ‘CI before AI’ approach to ensure that we don’t end up automating inefficient processes.”

Continuous improvement helps teams make sure the underlying work is worth scaling. That’s when a continuous improvement approach can help. It encourages practices like Gemba walks, Kaizen events, bowler cards, and monthly business reviews that allow our teams to understand where work gets stuck and where AI can help.

Continuous improvement keeps the council model grounded in real work. We’re applying it where the process is understood, the value is clear, and the outcome can be measured.

Scaling AI responsibly

Our compliance council encourages the application of Responsible AI, so our teams can move faster with confidence. As our AI work scales across Microsoft Digital, responsible AI has to connect directly to the same council ecosystem that guides strategy, enablement, data, process, and measurement. That connection helps teams understand what they’re accountable for before they build too far, too fast.

Our responsible AI work focuses on compliance, inclusiveness, fairness, transparency, reliability, privacy, security, and accountability. It’s grounded in the Microsoft Responsible AI Standard and supported by responsible AI champions who help teams apply those expectations in real development workflows.

This approach gives teams structure. It allows them to assess impact, identify risks, document decisions, and bring in the right reviewers. It also creates consistency, as more AI agents and solutions move from experimentation into enterprise use.

The goal is to enable AI project teams to move in the right direction with the right safeguards. Responsible AI gives the strategy council, the AI Center of Excellence, the data council, and product teams a shared standard for trust—to turn ambition into accountable execution. It also makes sure the AI systems we scale are worthy of the trust that employees, customers, and the company place in them.

Measuring our AI outcomes

Our councils choose the right AI work, support teams as they build, strengthen the data foundation, apply responsible AI, and improve processes before we scale. But we still need to answer the most important question: What changed because of the AI investment?

That’s why we have built a common value measurement framework across Microsoft Digital. Our teams use the framework to define expected value before they build. With it, they can establish a baseline, track results, and review what they learn with the right business and AI owners.

We organize AI value across six areas: Revenue impact, productivity and efficiency, security and risk management, employee and customer experience, quality improvement, and cost savings. Not every initiative needs to deliver value in every category. The point is to create a shared language that leaders and teams can use to compare investments, make tradeoffs, and understand progress.

Measurement also pushes us past simple savings claims.

If AI saves time, reduces cost, improves quality, or increases coverage, we want to know what happens next. Did teams reinvest that capacity? Did service improve? Did risk go down? Did quality increase?

AI accountability depends on that full loop. We define value, measure results, review progress, and adjust. Then we use what we learn to guide the next round of decisions.

Operating as one connected AI system

Our AI councils make a difference because each group has a different focus.

“What got us here won’t get us to where we need to go next. We started with broad experimentation—getting teams excited and building—but now we’re evolving as an organization to think about scale, alignment to business goals, and making sure our investments are driving the right outcomes.”

Myron Wan, principal group product manager, Microsoft Digital

Strategy assists us in choosing the right priorities. Enablement helps our teams to build with shared patterns. Data readiness gives AI systems a trusted foundation. Responsible AI allows us to move faster with confidence. Continuous improvement makes sure we’re improving the work before we automate it. Measurement tells us whether the investment changed anything meaningful.

Together, this system means we can operate AI as a business-driven enablement system.

“What got us here won’t get us to where we need to go next,” says Myron Wan, a principal group product manager in Microsoft Digital. “We started with broad experimentation—getting teams excited and building—but now we’re evolving as an organization to think about scale, alignment to business goals, and making sure our investments are driving the right outcomes.”

There’s more work ahead. We need to keep scaling enablement, improving data readiness, increasing high-value use cases, showcasing measurable impact, and tightening alignment across teams.

We also need to keep asking the hard questions: Where should we invest? Where are we reducing risk? Are we reinvesting the value that AI creates?

Our council-based model allows us to answer those questions with discipline. It helps us connect AI ambition to business outcomes and move from experimentation to repeatable enterprise value. And it provides a practical model that other IT organizations can adapt as they guide their own AI deployment.

The post Guiding our AI deployment with a set of employee councils appeared first on Inside Track.