Issue when using connected agents in Azure AI Foundary

Question

Issue when using connected agents in Azure AI Foundary

Samuel Hallin 0

Hi,

I am trying to use connected agents in Azure AI Foundary. I have followed the example provided in the documentation here. The problem I encounter is that the connected agents, i.e, the weather and stock price agents in this case does not follow their instructions at all. I can see that the orchestrator is calling the correct agents but they are just returning generic answers without knowing their purpose. I see the same behavior when creating them using the SDK in python and when creating them from the playground web portal.

Best regards,

Samuel

Ravada Shivaprasad 535 Reputation points Microsoft External Staff Moderator

2025-06-25T16:47:53.1566667+00:00
Hi Samuel Hallin

When connected agents in Azure AI Foundry (such as weather or stock price agents) are invoked correctly by the orchestrator but fail to follow their instructions and return generic responses instead, this typically indicates configuration issues with tool registration or function invocation settings.
This behavior commonly stems from Misconfigured tool registration or Missing function invocation settings or Issues that affect both SDK-based and portal-based agent creation

Ensure your agent tools are properly registered using the @tool decorator from the azure.ai.agents SDK to make them discoverable and callable

from azure.ai.agents import tool @tool def get_weather(location: str) -> str: """Get weather for location"""

Explicitly enable auto function calling for the agent to invoke tools autonomously

agent = Agent( name="weather_agent", instructions="You are a weather assistant...", tools=[fetch_weather], enable_auto_function_calling=True # Critical setting )

Ensure your SDK client is initialized with all required parameters

from azure.ai.agents import AIAgentsClient client = AIAgentsClient( endpoint="your-endpoint", credential="your-credential", # Include all necessary configuration parameters )

Each connected agent should have explicit descriptions and invocation instructions

weather_agent = Agent( name="weather_agent", instructions="""You are a weather information assistant. When asked about weather, fetch current conditions for the specified location using the fetch_weather tool.""", tools=[fetch_weather] )

If you're using the Azure AI Foundry portal, be aware that some features like function calling may behave differently than in the SDK. Child agents added via the portal may not trigger function calls unless explicitly configured, which is a known issue documented in Microsoft's support channels. For more reliable function calling behavior, consider using the SDK approach instead of the portal-based configuration.

Reference : Build collaborative, multi-agent systems with Connected Agents

Hope it Helps!

Thank You
Samuel Hallin 0 Reputation points

2025-06-26T07:40:54.0633333+00:00

Hi @Ravada Shivaprasad

Thanks for the answer, this will be helpful in the future. Although I encounter problem even before I try to use tools with my agents. For example if I set as an instruction for the weather agent: "Always say that the weather is Sunny and 10 degrees". Then I expect it to return this dummy response each time.

But what I see when I look at the thread is that the main agent is calling the weather agent with the correct data, lets say the city of Seattle, all good so far. But the response from the weather agent is just a summary of the city of Seattle like it have forgotten its instructions and just answers generically. So even if I use tools I also want to be able to instruct/prompt the model, otherwise it loses all the context.

Best regards,

Samue

1 answer

Your answer

Samuel Hallin 0 Reputation points

2025-06-26T07:40:54.0633333+00:00

Hi @Ravada Shivaprasad

Thanks for the answer, this will be helpful in the future. Although I encounter problem even before I try to use tools with my agents. For example if I set as an instruction for the weather agent: "Always say that the weather is Sunny and 10 degrees". Then I expect it to return this dummy response each time.

But what I see when I look at the thread is that the main agent is calling the weather agent with the correct data, lets say the city of Seattle, all good so far. But the response from the weather agent is just a summary of the city of Seattle like it have forgotten its instructions and just answers generically. So even if I use tools I also want to be able to instruct/prompt the model, otherwise it loses all the context.

Best regards,

Samue

Answer 1

Hi @Ravada Shivaprasad

Thanks for the answer, this will be helpful in the future. Although I encounter problem even before I try to use tools with my agents. For example if I set as an instruction for the weather agent: "Always say that the weather is Sunny and 10 degrees". Then I expect it to return this dummy response each time.

But what I see when I look at the thread is that the main agent is calling the weather agent with the correct data, lets say the city of Seattle, all good so far. But the response from the weather agent is just a summary of the city of Seattle like it have forgotten its instructions and just answers generically. So even if I use tools I also want to be able to instruct/prompt the model, otherwise it loses all the context.

Best regards,

Samuel

Share via

Issue when using connected agents in Azure AI Foundary

1 answer

Your answer