Lean Into Model Strengths: Express Actions as Code Over Opaque Tool Calls

cedric_chee · x · 2026-08-06

A developer points out that when building AI applications, it's crucial to leverage what large models are naturally good at. Because models have seen vastly more code than tool calls during training, they inherently want to write code.

He suggests that AI harnesses should express actions as code rather than opaque tool invocations. This approach works with the model's strengths instead of against them, optimizing performance.

Original post →

More from coding & agent

coding & agent channel →