From Demos to Production: What Microsoft Foundry Now Is
Microsoft Foundry is an AI app and agent factory that gives enterprises a single runtime, tooling stack, and governance model so AI agents can move from experimental demos into production deployment under consistent policies, observability, and infrastructure. At Build, Microsoft positioned Foundry as “the place where AI agents move from experiments to production systems,” focusing on reliable operations more than headline model features. The platform now spans hosted agent infrastructure, memory, grounding, models, and shared observability across every agent. This shift reflects a belief that the real competition in enterprise AI agents is reliability and governability rather than raw capability. For teams that have struggled to keep prototypes alive under real load, real data, and real compliance requirements, Foundry’s updated runtime and governance features aim to replace ad‑hoc stacks with a single managed environment for AI agents in production.
Microsoft Foundry Runtime: Hosted Agents Without Rewrites
The new Microsoft Foundry runtime centers on hosted agents in the Foundry Agent Service, which provides a managed environment where each session runs in its own sandbox with compute, memory, and durable filesystem access. Enterprise teams can bring agents built with Microsoft Agent Framework, GitHub Copilot SDK, LangGraph, or other SDKs and deploy them without rewriting code. Foundry exposes a stateful Responses API for OpenAI‑style interactions and an invocations protocol for pass‑through calls where developers control request and response formats. This flexibility helps teams preserve existing orchestration while standardizing on a single AI agents production deployment platform. Long‑running autonomous agents gain durable state, and routines, now in public preview, can run agents on schedules for tasks such as overnight ticket triage or daily reporting, turning agents into reliable background workers rather than one‑off chats.
Toolboxes, Memory, and Knowledge: Making Agents Enterprise-Ready
Toolboxes in Microsoft Foundry, now in public preview, attack the tool governance problem by giving each agent a single managed endpoint for tools, skills, Model Context Protocol clients, and enterprise data integrations. Instead of hard‑coding tools into every agent, teams register them once, manage access centrally, and let agents discover what they need at runtime. Skills are versioned in a project-scoped catalog and surfaced via MCP, while tool search narrows the tools exposed per task to keep context windows small and outputs focused. Foundry treats memory as a platform function, with procedural, user, and session memory hosted in the Foundry Agent Service. According to Nick Brady, early Tau benchmarks show procedural memory delivering “7 to 14 percent absolute success rate gains at near baseline cost,” helping agents learn how to carry out work across runs using a managed, inspectable store.
Governance and Logic Apps Automation: One Platform, Managed Policies
Governance is the other pillar of the updated Microsoft Foundry runtime. Microsoft introduced ASSERT, an open-source framework that converts written policies into measurable evaluations for agent behavior, enabling regression tests that align with enterprise AI governance rather than static benchmarks. A complementary Agent Control Spec defines how teams describe what agents are allowed to do, turning policy into something testable and automatable. On the workflow side, Logic Apps Automation integrates workflows, AI agents, knowledge services, and model access into a managed SaaS experience on auto.azure.com. Developers can orchestrate agents alongside existing processes without stitching together multiple tools, and they gain shared observability and policy enforcement across every agent. Together, these features turn Foundry into a standard platform where enterprise AI governance, runtime behavior, and production workflows are managed in one place instead of scattered across custom stacks.
