I completely agree; I don't know how to code at all. For the agent-based products I'm building now, I delegate code review to a dedicated "reviewer" agent, and then the primary agent makes further modifications based on the errors found—repeating the process until every issue is resolved. To my surprise, this approach not only achieved my initial goals but also resolved other underlying issues within the workflow, far exceeding my expectations. This way, AI development isn't limited by my own input and output capabilities.

https://x.com/unclebobmartin/status/2080257779395154409?s=46&t=hMgsENEOQpNdIVfz56TpfA