#sandbox

Letting Agents Run, Safely

Part three of a series on agent harnesses. Part one introduced the model-plus-harness idea, and part two split work across subagents. This piece is about the layer where agents stop just talking and s