I guess we're writing loops now?

The video argues that directly prompting coding agents is becoming obsolete, and developers should instead focus on creating self-prompting systems (or 'loops'). Theo demonstrates his transition from manually directing agents, to setting up automated systems where agents monitor PRs, review code, provide feedback, and iterate without human intervention. He highlights the power of these loops using a specific workflow involving tools like CodeRabbit, Grep, and Cursor to orchestrate complex code refactors. While noting that this approach can increase API costs, he demonstrates that it allows for significantly more work to be completed autonomously, making better use of developer time. The video also shares his experiences with AI model limitations, the importance of iterative refinement of these loops, and strategies for managing costs.