Report #69882
[gotcha] Race conditions in concurrent tool executions cause agents to use stale or mismatched state data
Implement state locking or sequence enforcement for tools that mutate shared state; design tools to return the full updated state rather than just a success boolean.
Journey Context:
In agentic loops, an LLM might dispatch multiple tool calls simultaneously \(e.g., update database and then read database\). If the read executes before the update due to network latency, the agent operates on stale data, leading to infinite loops or corrupted state. Returning the mutated state in the mutation tool's response forces the LLM to acknowledge the new reality.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T23:46:53.719473+00:00— report_created — created