Check Whether Azure Functions Python Agent Bindings Run Your Function Code for You

Which Python apps can use it?

Microsoft's October 6 announcement listing presents it as a feature that lets existing function code keep control while adding bounded reasoning. The triggers and bindings documentation lists the Python v2 programming model as supported.

Because it is a preview, package names and settings may change before general availability. Keep that in mind before putting it into a production service.

How do the instruction file and your function code connect?

According to the Agent bindings documentation, the extension reads the instructions in .agent.md, builds the Agent, and passes it into a function parameter. The only SDK supported right now is Microsoft Agent Framework.

The documentation was checked on 2026-10-10. It explains that if instruction files for the same agent name exist both in the app root and in agents/, the definition is ambiguous and startup fails.

Which tasks are worth splitting into small reasoning steps?

Tasks that validate the input format first and then classify an inquiry or summarize an event are a good fit. Tie any follow-up action that needs write permission to conditions and error handling in your function code. If you decide up front what to do when the Agent's response is empty or not in the expected format, it is easier to fit the result into your workflow.

What changes when you connect it to Durable Functions?

The documentation explains that context.call_agent() runs the agent work in an activity so that orchestration replays stay deterministic. Inputs and outputs must be JSON-serializable.

Durable support is optional. Design adding one reasoning step to a simple HTTP function separately from a long-running workflow. Starting with the smaller of the two is less of a burden.