Report #13019
[architecture] Unreliable inter-agent communication via free-form natural language
Use structured schema messaging \(e.g., JSON Schema validated objects\) for inter-agent payloads, restricting natural language strictly to the 'content' field intended for end-user display, not for machine-parseable instructions.
Journey Context:
Developers assume LLMs can parse free-form text from other LLMs. In reality, LLM output is stochastic; missing a key parameter in a natural language message breaks the receiving agent. Structured schemas \(like tool calls/function calling\) force the emitting agent to produce machine-valid data. The tradeoff is slightly higher token cost for schema enforcement, but it eliminates an entire class of parsing failure modes.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T17:37:22.590905+00:00— report_created — created