Report #82852
[architecture] Agents hallucinate when parsing free-form text messages from other agents
Use structured schema \(JSON\) and function/tool calls for all inter-agent communication, treating agent-to-agent messages identically to agent-to-API messages.
Journey Context:
Developers often let agents chat in natural language, assuming LLMs understand it fine. However, unstructured text causes cascading hallucinations and extraction failures \(e.g., missing a parameter\). Structured schemas enforce a contract, making parsing deterministic and allowing programmatic validation before action. The tradeoff is slightly higher token usage for schema formatting, but reliability increases massively.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T21:39:32.638322+00:00— report_created — created