Report #59272
[architecture] Agent treats memory as an append-only chat log, failing to update its internal state representation of the user or task
Separate episodic memory \(what happened\) from semantic memory \(what is true\). When a user provides a profile update, extract the semantic fact and update the user's entity state, rather than just appending the message to episodic memory.
Journey Context:
Append-only logs are easy to build but terrible for reasoning. If the agent needs to know a user's email, searching episodic memory might return 'I think my email is X' instead of the authoritative fact. By maintaining a structured semantic state alongside the episodic log, the agent has a source of truth that doesn't require inference.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T05:58:39.641781+00:00— report_created — created