Microsoft Foundry makes scheduled and event-triggered routines generally available

A routine invokes an agent at a chosen time, on a recurring schedule or when a supported external event arrives, supplying the starting mechanism for work the agent is already configured to do.

The initial event connections cover opened or closed GitHub issues and new messages in Microsoft Teams channels. Foundry keeps the trigger, connection and run record with the agent’s project. For a team assembling a daily research digest or issue-triage assistant, that can replace a separate piece of scheduling infrastructure with a managed entry point.

A routine has a deliberately narrow shape: one trigger and one action invoking one prompt or hosted agent. Its usefulness is therefore clearest when the task already works and simply needs to start automatically. Branching decisions, multiple agents and approval logic still require an appropriate workflow design; enabling a timer does not supply them.

Identity is an equally consequential choice. The agent’s identity is the default; using the routine creator’s identity requires an explicit choice at creation. That choice determines whose permissions a run carries. It should follow the intended task, especially where tools can read private information or change an external system.

The documentation offers a practical first test: run a small routine manually, inspect its input, status and output, and only then enable the schedule. Begin with a result someone can review, such as a draft digest. The operational gain is a clearer, inspectable route from an event to a run. General availability does not demonstrate that the agent’s conclusions are correct or that its proposed actions are appropriate.