How to use an agent in a workflow step with structured output
I want a workflow step that calls an LLM and returns a structured response (validated against a Zod schema).
Right now the only way I know is to wrap the agent call inside a custom execute function. That works, but the downside is the workflow graph won’t show the agent call as its own step — it just looks like a generic custom step.
Is there a way to make the agent itself be the workflow step, while still enforcing structured/Zod output?
Example of what I’m doing now:
This works, but since the agent call is hidden inside execute, the workflow graph doesn’t visualize the agent step. How can I wire it up so the graph shows the agent call itself as the step, while still getting structured output?
Right now the only way I know is to wrap the agent call inside a custom execute function. That works, but the downside is the workflow graph won’t show the agent call as its own step — it just looks like a generic custom step.
Is there a way to make the agent itself be the workflow step, while still enforcing structured/Zod output?
Example of what I’m doing now:
This works, but since the agent call is hidden inside execute, the workflow graph doesn’t visualize the agent step. How can I wire it up so the graph shows the agent call itself as the step, while still getting structured output?
