Microsoft Build backs governance startups

Microsoft is using its Build 2026 startup lineup to spotlight a different part of the AI market: the software that governs, observes and connects agents once companies try to run them inside real businesses. In a May 21 post, Microsoft for Startups said the 11 companies selected for this year’s Build cohort span developer tooling, AI infrastructure and physical AI, and share “a bias toward solving real enterprise problems rather than interesting-sounding ones.” The event runs June 2-3 in San Francisco and online, according to the Build conference site. Microsoft’s own framing is less about new foundation models than about what it takes to move AI from prototype to production. ### Why does this cohort look different from last year’s AI startup story? Microsoft for Startups said on May 21 that “the conversation has shifted” from whether founders should build with AI to “how to make AI work in production, at scale, across real enterprise systems.” The company said that change shaped this year’s Build cohort and its programming. The same post said the featured startups are tackling “the hard, unglamorous problems between a good idea and a system that actually runs,” including authenticating agents, understanding legacy code safely and measuring whether AI tools are improving engineering work. Microsoft described the infrastructure layer as the place where enterprises have discovered gaps around hallucinations, agent drift, sensitive data and rising costs. (microsoft.com) ### What kinds of problems is Microsoft elevating? Microsoft’s Build agenda for startups is organized around production deployment, cost control and scaling, according to a May 5 Microsoft for Startups post listing recommended sessions. The sessions include “Take LLMs from prototype to production on AKS,” “Postgres and OpenAI – a match made in Azure,” and “The honest practitioner’s take on agentic AI on Kubernetes.” (microsoft.com) Microsoft said on its Build site that the conference will focus on “real code and real systems” built to scale AI. That language matches the startup post’s emphasis on compute, observability and data systems that can hold up under enterprise conditions. ### Where does governance fit into Microsoft’s broader product push? Vasu Jakkal, Microsoft’s corporate vice president for security, wrote on March 9 that Microsoft Agent 365 is a “control plane for agents” designed to help organizations observe, govern and secure agents across an enterprise. (microsoft.com) She said customers were asking how to track agents, understand what they do, verify access and prevent sensitive data leaks. (build.microsoft.com) Microsoft said Agent 365 includes an agent registry and gives IT, security and business teams visibility into managed agents in their environment. In the same post, Avanade Chief Technology and Information Officer Aaron Reich said the product gave the company “real visibility into agent activity,” the ability to govern agent sprawl and control resource usage. (microsoft.com) ### Why is observability showing up beside governance and data? Microsoft Security researchers wrote on March 18 that observability is “one of the foundational security and governance requirements” for AI systems in production. They said agentic systems can interact with sensitive data, call external APIs, initiate workflows and collaborate with other agents, creating blind spots if companies cannot see what those systems are doing. (microsoft.com) The researchers said traditional software telemetry is not enough because AI systems are probabilistic and can take less predictable paths. They argued that organizations need new signals and telemetry to understand and govern AI behavior, including cases where one agent passes poisoned content to another. (microsoft.com) ### What does that say about where money and partnerships are going? Microsoft said many of the 11 Build startups are members of its Pegasus Program or are backed by M12, the company’s venture arm, and all are available through Microsoft’s marketplace channels. The company also said the startups are clustered around the engineering decisions that determine whether AI “ships, scales, and holds up under real conditions.” (microsoft.com) Microsoft Build opens on June 2 in San Francisco, where the company said the AI infrastructure ecosystem is dense and close to enterprise buyers. The startup track will continue through the conference with sessions on production architectures, cost-optimized deployment and agentic systems, according to Microsoft’s event materials. (microsoft.com 1) (microsoft.com 2)