Report #92736
[agent\_craft] Agent makes unnecessary sequential calls when dependencies don't exist
Use dependency-graph analysis: if tool calls have no data dependencies \(independent parameters\), batch them in a single request using parallel\_tool\_calls \(OpenAI\) or multiple tool\_use blocks \(Anthropic\); for dependent chains \(B needs result of A\), force sequential. Explicitly mark independent batches with tags in reasoning.
Journey Context:
Many agents default to sequential execution 'just to be safe,' doubling or tripling latency for independent operations \(e.g., reading two unrelated files\). The fix requires analyzing the parameter graph before calling. OpenAI supports explicit parallel\_tool\_calls flag; Anthropic allows multiple tool\_use blocks in one assistant message. The risk is parallelizing dependent calls \(race conditions\). The tag in reasoning forces the agent to explicitly declare independence.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T14:14:49.769283+00:00— report_created — created