Microsoft has expanded Copilot Studio from an agent builder toward a single environment for apps, deterministic workflows and agents. Apps are now in public preview, allowing makers to describe an application in natural language, refine the generated result conversationally and access underlying code when deeper customization is needed. Copilot Managed Runtime, also in public preview, provides Microsoft-operated hosting, identity, governed data access, lifecycle management and Git-backed versioning.
For agents using the GitHub Copilot harness, new hooks in preview add deterministic logic at lifecycle points such as session start, tool calls and errors. A hook can load required context, inspect or block a tool call, transform a result or capture consistent telemetry. This gives teams a way to enforce actions that must always happen while leaving the agent free to reason elsewhere.
The Foundry IQ integration is now generally available, enabling reusable enterprise knowledge bases across Copilot Studio agents with citations and private-network connectivity. Evaluation capabilities are also expanding to measure task completion, tool accuracy, safety and latency, while custom graders can be reused across agents and test sets. The Review panel is generally available and surfaces blockers, warnings and policy restrictions before publication.
The update is significant because it combines probabilistic agent behavior with deterministic workflow controls and governance in one product surface. However, maturity varies: app building, Managed Runtime and hooks remain previews, while Foundry IQ integration and the Review panel are generally available. Microsoft's claims about improved predictability and enterprise readiness describe product intent; this release does not itself establish independent productivity or reliability gains.