Microsoft Foundry Workflow Issue - Agent based on Claude Models

Question

Hi all,
I have a Workflow in Foundry that invokes an Agent (Model: Claude-Sonnet-4.5) to generate Data Pipeline JSON templates. When i invoke the agent from workflow, the agent only shows me the notification message and shuts without actually showing the response output.

I have tested this agent (Model: Claude-Sonnet-4.5) separately and I am able to see the responses after 10/15 seconds. although, it occasionally does fail to generate the response. I wanted to know if there are any response timeouts when it comes to calling agents from the workflow or is it a limitation with Claude models? When I have tested the same with GPT I get the response.

The reason behind inclining more towards Claude LLMs instead of GPT is that, Claude is more accurate when comes to generating codes and I have also seen proof of that.

Answer 1

Hey AyusmanBasu-6332,

Thank you for reaching out on the Microsoft Q&A.

It looks like you're running into an issue where the Claude-Sonnet-4.5 agent isn't returning responses as expected when invoked from your workflow in Foundry. This can definitely be frustrating, especially since you've tested the agent separately and received responses.

Here's a breakdown of what might be causing the issue:

1. Response Timeouts: While the documentation doesn’t explicitly mention timeouts for agent calls within workflows, it's important to note that agents may have different performance characteristics depending on the usage context. If you’re experiencing timeouts or failures only with Claude models in a specific workflow setup, it might be worth checking if there are any configuration differences or constraints compared to when you're testing it separately.
2. Model Limitations: You mentioned that Claude models sometimes fail to generate responses while GPT models work as expected. This could be due to specific limitations or performance characteristics inherent to the Claude model itself. Ensure your model deployment is correct and that you’re following the best practices for using Claude within Foundry.
3. Configuration and Permissions: Make sure that your workflow has the necessary permissions configured correctly, particularly if you are using features that require special roles like Cognitive Services OpenAI Contributor for accessing the Azure OpenAI resource. Double-check that there are no access limitations in your workflow’s settings. CLI snippet for permissions check:

      az role assignment list --assignee <your-user-id>
   

4. Network Configuration: If your projects are operating within a virtual network (VNET) or have specific networking setups, ensure that all resources are in the same region and that your VNET configurations are correct. Mismatched regions or incorrect networking setup can lead to unexpected behavior.
5. Monitoring and Diagnostic Logs: Utilize Azure Monitor to track any logs or error messages that might be generated when the agent fails to respond as expected. This could help provide additional context for the issue.

References

• Use your own resources in the Azure AI Foundry Agent Service
• How to use a virtual network with the Azure AI Foundry Agent Service
• Model and region support
• Quotas and limits
• Set up required permissions
• Deploy and use Claude models in Microsoft Foundry (preview)

I hope this helps clarify some potential areas to investigate! Do let me know if you have any further queries.

---

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thankyou!