Report #7342
[architecture] Agents failing to parse free-form text messages in multi-agent pipelines
Use structured schemas \(JSON/Pydantic\) for inter-agent messaging, reserving natural language only for final user-facing output or internal reasoning.
Journey Context:
LLMs are good at generating text but unreliable at parsing unstructured text into strict action parameters. Free-form chat causes cascading parsing errors. Structured schemas \(like function calls\) force the LLM to emit machine-readable state, trading a bit of expressive flexibility for deterministic routing and execution.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T02:33:58.682849+00:00— report_created — created