{"id":1155354,"date":"2025-12-04T04:13:11","date_gmt":"2025-12-04T12:13:11","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-project&#038;p=1155354"},"modified":"2026-02-02T09:22:47","modified_gmt":"2026-02-02T17:22:47","slug":"agentguard-early-warning-and-routing-for-predictable-agenticai-on-azure","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/agentguard-early-warning-and-routing-for-predictable-agenticai-on-azure\/","title":{"rendered":"AgentGuard: Early-Warning and Routing for Predictable AgenticAI on Azure"},"content":{"rendered":"<section class=\"mb-3 moray-highlight\">\n\t<div class=\"card-img-overlay mx-lg-0\">\n\t\t<div class=\"card-background  has-background- card-background--full-bleed\">\n\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"1920\" height=\"721\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1.jpg\" class=\"attachment-full size-full\" alt=\"background pattern\" style=\"\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1.jpg 1920w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1-300x113.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1-1024x385.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1-768x288.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1-1536x577.jpg 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1-1600x600.jpg 1600w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/11\/AgentGuard_-Early-Warning-and-Routing-for-Predictable-Agentic-AI-on-Azure_Banner-1920x721-1-240x90.jpg 240w\" sizes=\"auto, (max-width: 1920px) 100vw, 1920px\" \/>\t\t<\/div>\n\t\t<!-- Foreground -->\n\t\t<div class=\"card-foreground d-flex mt-md-n5 my-lg-5 px-g px-lg-0\">\n\t\t\t<!-- Container -->\n\t\t\t<div class=\"container d-flex mt-md-n5 my-lg-5 \">\n\t\t\t\t<!-- Card wrapper -->\n\t\t\t\t<div class=\"w-100 w-lg-col-5\">\n\t\t\t\t\t<!-- Card -->\n\t\t\t\t\t<div class=\"card material-md-card py-5 px-md-5\">\n\t\t\t\t\t\t<div class=\"card-body \">\n\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\n\n<h1 class=\"wp-block-heading\" id=\"agentguard-early-warning-and-routing-for-predictable-agenticai-on-azure\">AgentGuard: Early-Warning and Routing for Predictable Agentic<br>AI on Azure<\/h1>\n\n\n\n<p><\/p>\n\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n<\/section>\n\n\n\n\n\n<p>This project introduces AgentGuard, a monitoring and routing system designed to improve reliability and cost-efficiency in Azure-based agent workflows. By analysing early trajectory signals\u2014such as reasoning patterns and tool usage\u2014within the first 10\u201330% of an agent\u2019s execution, AgentGuard predicts task outcomes and dynamically halts or re-routes agents to prevent failures. Calibrated with ADeLe for demand-aware evaluation, the system aims to optimise the Reliability\u2013Latency\u2013Cost frontier across diverse benchmarks, including software engineering and cybersecurity tasks. Expected outcomes include integrated Azure routing capabilities, diagnostic insights into team composition (optimized on agent psychometrics, collective agent behavior\/ intelligence, neural evolutions, and evolutionary dynamics), and practical strategies for building more predictable, trustworthy agentic AI.<\/p>\n\n\n\n<p>This research is conducted via&nbsp;<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/academic-program\/agentic-ai-research-and-innovation\/\">The Agentic AI Research and Innovation&nbsp;<\/a>(AARI) Initiative which focuses on the next frontier of agentic systems through&nbsp;<em>Grand Challenges<\/em>&nbsp;with the academic community and Microsoft Research.<\/p>\n\n\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>This project introduces AgentGuard, a monitoring and routing system designed to improve reliability and cost-efficiency in Azure-based agent workflows. By analysing early trajectory signals\u2014such as reasoning patterns and tool usage\u2014within the first 10\u201330% of an agent\u2019s execution, AgentGuard predicts task outcomes and dynamically halts or re-routes agents to prevent failures. Calibrated with ADeLe for demand-aware [&hellip;]<\/p>\n","protected":false},"featured_media":1155709,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","footnotes":""},"research-area":[13556],"msr-locale":[268875],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1155354","msr-project","type-msr-project","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-locale-en_us","msr-archive-status-active"],"msr_project_start":"","related-publications":[],"related-downloads":[],"related-videos":[],"related-groups":[],"related-events":[],"related-opportunities":[],"related-posts":[],"related-articles":[],"tab-content":[],"slides":[],"related-researchers":[{"type":"guest","display_name":"&Aacute;lvaro David G&oacute;mez Ant&oacute;n","user_id":1161258,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Keno Harada","user_id":1161260,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Jos&eacute;  Hern&aacute;ndez-Orallo","user_id":1157487,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Kazuya Horibe","user_id":1157643,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Ilya Horiguchi","user_id":1161262,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Kexin  Jiang-Chen","user_id":1157641,"people_section":"Section name 0","alias":""},{"type":"user_nicename","display_name":"Haotian Li","user_id":43593,"people_section":"Section name 0","alias":"haotianli"},{"type":"guest","display_name":"Ryutaro Mori","user_id":1161261,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Yael Moros Daval","user_id":1157485,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Peter Romero","user_id":1157483,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Daniel Romero-Alvarado","user_id":1157484,"people_section":"Section name 0","alias":""},{"type":"guest","display_name":"Kotaro Sakamoto","user_id":1161259,"people_section":"Section name 0","alias":""},{"type":"user_nicename","display_name":"Beibei Shi","user_id":42162,"people_section":"Section name 0","alias":"besh"},{"type":"guest","display_name":"Wataru  Toyokawa","user_id":1157642,"people_section":"Section name 0","alias":""},{"type":"user_nicename","display_name":"Xing Xie","user_id":34906,"people_section":"Section name 0","alias":"xingx"}],"msr_research_lab":[],"msr_impact_theme":[],"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1155354","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-project"}],"version-history":[{"count":7,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1155354\/revisions"}],"predecessor-version":[{"id":1158814,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-project\/1155354\/revisions\/1158814"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1155709"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1155354"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1155354"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1155354"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1155354"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1155354"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}