Theo Browne argues that as coding agents generate more code, developers need to invest in the environment around that workAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction.. His gardening analogy emphasizes improving the conditions in which changes are made and removing recurring sources of friction.
Theo Browne recounts working across product and platform teams at Twitch and later adapting his own stack as a small team grew. He describes using feedback from product development to improve the shared tools and conventions that support it.
Theo Browne discusses trimming tests he considered unhelpful, investigating why agents add themHuman judgment in AI is the accountable interpretation and decision-making people contribute when setting goals, evaluating evidence, and managing consequences., and adding a focused check for a recurring performance regressionEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. These examples are presented as his experience and judgment, not as a general recommendation to remove tests.
Watch on YouTube
