Building and customizing solutions using Microsoft 365 Copilot APIs and tools
Something went wrong is a generic agent failure, not a single root cause. In this scenario, the most likely causes are flow response/schema problems between the agent and the flow.
Most relevant causes to check:
- The flow response no longer matches what the agent expects
- If the flow input or output schema changed and the agent action wasn’t refreshed, the agent can fail even though the flow runs.
- In channels such as Teams, this commonly appears as
FlowActionBadRequestand is caused by input/output schema mismatches or null/missing values.
- The flow isn’t returning the expected output to the agent
- If no output is returned, or an expected output parameter is missing, the agent can fail with flow-related errors.
- Documented examples include missing output values or output schema mismatches.
- The flow is configured to respond asynchronously instead of in real time
- Agent flows must return values synchronously.
- If Asynchronous response is turned on in Respond to the agent, the agent can fail and show:
Something unexpected happened. We're looking into it. Error code: 3000.
- A null or unsupported parameter type is being passed
- Supported flow parameter base types are Text, Boolean, and Numbers.
- Null or missing inputs can also trigger failures.
- Teams or channel conversation state is stale
- Teams can cache an older version of the agent conversation, so the error can continue even after the flow was fixed.
Actionable troubleshooting path:
- In Copilot Studio, open the topic that contains the flow action.
- On the Action node, select the ellipsis (...) > Refresh.
- Verify the input and output parameters in Copilot Studio exactly match the parameters in Power Automate.
- Save and republish the agent.
- Open the flow and check the Respond to the agent action:
- Go to Settings
- Confirm Asynchronous response is Off
- Save the flow
- Check that every value passed to the flow is valid:
- No null required inputs
- Only supported base types: Text, Boolean, Numbers
- If the flow appears to run but the agent still does not pick up the response, verify the flow actually returns every expected output parameter.
- In Teams, start a fresh conversation by typing start over or goodbye.
- If the issue persists in Teams, enter
/debug clearstateto clear the conversation state and reload the agent state. - Test again in Test your agent after each change.
If the failure happens mainly in Test your agent and the chat stops responding after the first message, check whether a corporate firewall or proxy is buffering the /subscribe Server-Sent Events connection. That can make the panel appear unresponsive even when activity exists in the transcript.
What the message means in practice:
- It usually indicates the agent could not successfully consume the flow result, not necessarily that the flow never started.
- For this specific symptom — flow runs, but the agent cannot pick up the response — the strongest documented causes are:
- stale schema in the action node,
- missing flow outputs,
- null/mismatched input values,
- asynchronous response enabled.