Report #29181
[counterintuitive] The largest, most capable model should execute all agent steps
Use an orchestrator/router pattern: a large model for planning and complex reasoning, and a fast, small model for routine tool execution and formatting.
Journey Context:
Using frontier models for every step \(including formatting a simple CLI output or reading a trivial file\) is slow and expensive. Smaller models are highly capable of executing well-defined tool calls if the parameters are already determined. Separating the 'planner' from the 'executor' optimizes for both latency and cost without sacrificing the quality of the final output.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T03:22:29.421431+00:00— report_created — created