I guess we're writing loops now?

Coding agents should transition from manual, human-prompted tasks to autonomous, self-correcting loops. By designing systems where agents audit their own work, manage sub-threads, and handle multi-stage pull requests, developers can move beyond the "copy-paste" cycle toward dynamic, problem-specific orchestration. This approach replaces static, hard-coded personas with flexible workflows that adapt to the task at hand. While looping significantly increases token usage and costs, the resulting productivity gains—such as executing complex refactors or monitoring repository issues overnight—often justify the investment. Developers should identify their own repetitive post-prompting steps, such as testing, committing, or reviewing, and delegate these to the agent. Ultimately, the goal is to minimize human intervention, reserving human effort for high-level decision-making while agents handle the iterative, execution-heavy lifecycle of software development.