About
I am a Principal Research Software Engineer in Microsoft CoreAI and an alum of the Human-AI eXperiences (HAX) team at Microsoft Research, where I focus on AI agents and the research-to-product transition. My interests span agent frameworks, human-AI interaction patterns for agentic applications, developer experiences, and applied machine learning. My research has been published at conferences including EMNLP, ACL, CHI, NAACL, AAAI, and in Communications of the ACM, with multiple best paper awards. My work has been featured in outlets such as the Wall Street Journal and VentureBeat.
At Microsoft Research, I was a founding/core developer on AutoGen (opens in new tab), co-led the AutoGen 0.4 architectural rewrite, and designed the AgentChat API shape that subsequent agent frameworks (Google ADK, OpenAI Agents SDK, others) converged on. I created AutoGen Studio (opens in new tab) (EMNLP 2024 (opens in new tab)), the first visual workflow editor for multi-agent systems. I am a co-author on Magentic-One (opens in new tab) (state-of-the-art on the GAIA benchmark at release) and Magentic-UI (opens in new tab) (CHI 2026). I led the design of LIDA (opens in new tab) (ACL 2023 (opens in new tab)) for LLM-driven data visualization — approaches that were adopted into Microsoft Excel, Fabric, PowerBI, and Project Sophia (opens in new tab). I also developed evaluation metrics used in production model-selection for GitHub Copilot (ACL 2023 paper (opens in new tab), patent (opens in new tab)).
In Microsoft CoreAI, I contributed to the foundational architectural design for the Microsoft Agent Framework (opens in new tab) and built DevUI (opens in new tab) (~1M monthly downloads), MAF’s visual development environment for inspecting and debugging agents and multi-agent workflows.”. I currently serve as engineering architect on Agent Optimizer (opens in new tab) in Foundry Agent Service — a managed service that automatically improves AI agents on Azure. I am the author of Designing Multi-Agent Systems (opens in new tab) (398 pages, Amazon Best Seller, #1 New Release across multiple AI categories) and co-inventor on 8 US patents (granted and pending) spanning AI safety (opens in new tab), code-generation evaluation (opens in new tab), automated visualization (opens in new tab), and ML evaluation (opens in new tab).
Additional details on my work and interests can be found on my personal website – https://victordibia.com. (opens in new tab)