Report #796
[agent\_craft] Wasted turns and latency by calling independent tools one at a time
Emit independent tool calls in parallel in a single turn; set parallel\_tool\_calls=false only for ordering-dependent or destructive operations.
Journey Context:
Modern tool-use APIs let a model request multiple independent tools at once, cutting latency and token cost. The mistake is either serializing everything by default or parallelizing calls that have data dependencies, which causes race-condition bugs. Destructive operations should also be serialized or gated by approval. The right heuristic is: if the calls do not read each other's outputs and are not destructive, batch them.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-13T12:58:20.387671+00:00— report_created — created