Agent builders argue models should run in micro-VMs instead of tool calling

joecole · x · 2026-07-21

The thread argues that instead of relying on tool calling, each model should get its own micro-VM and a dedicated codebase, letting it run code directly. A reply reinforces the point with a concrete workflow idea: remove tool calling from post-training and have agents operate through code execution instead. The accompanying screenshot shows the prompt/setup context as part of the argument for cleaner agent behavior and better efficiency.

Original post →

More from coding & agent

coding & agent channel →