[BUG] Structure Output from Preventing tool invocation

[mekjr1]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

0.1.9

Python Version

3.12

Operating System

Ubuntu 24.04.1 LTS

Installation Method

pip

Steps to Reproduce

1. Simply create an agent with tools.

Expected Behavior

Normal flow, tools being called and then MyCustom model object as the output

Actual Behavior

Tools are not being called, but the output is MyCustom model object

Additional Context

From the looks of the outputs, its seems like the structured_output feature is registering a named after the response Model type. and that is the only tool being called

Possible Solution

No response

Related Issues

No response

[mekjr1]

@theagenticguy can you please assist? :)

[theagenticguy]

This is expected, we debated the dev experience of using structured output and chose an approach similar to other OSS frameworks. What it means is, you should use structured output in one of two ways:

1. First, call you agent normally via your_agent() one or more times (which would also potentially use your tools), then simply call your_agent.structured_output(your_model). This approach will use all the conversation history/context from the prior call you made. T
2. Use any logic you please to format a string (e.g. read a file, other LLM calls, etc ), and then you use agent.structured_output(output_model, "your string")

In other words, structured output itself isn't a tool, it's a method that effectively forces structured output when you call it. Under the hood, it starts a separate LLM invoke with only the structured output tool.

All that being said, would love to understand your expected experience and if you can share a code sample of what you are trying to accomplish, happy to help and take feedback

[dbschmigelski]

Marking this as a duplicate of #309 .

This appears to be a related issue where there is a customer expectation that structured output functions like agent.structured_output(output_model, {prompt) where the structured_output call initiates an event loop which structures the result of a "run"

Instead the flow now is

agent = Agent(tools=[...])
result = agent("Do something with my tools")
structured_result = agent.structured_output(output_model, result)

where structured_output "structures" the result of an event loop, it does not initiate a new one.