The author shares their approach to engaging with AI coding agents, specifically using Claude Code for project planning. They emphasize the importance of providing inline feedback directly in the plan’s markdown file. By leaving comments as they review the plan, the author finds they can address issues more effectively than using the chat interface of the AI. This method keeps them actively involved in the planning process, avoiding the pitfalls of disengagement that can lead to mistakes and oversights. The process starts with generating an initial plan through standard prompts. After obtaining the plan, they open the markdown file in their editor and leave feedback in the form of COMMENT: lines. The author describes this as akin to a code review, where they can add comments at their convenience, even within tables or other sections without worrying about strict markdown syntax. They illustrate this with a practical example where they had to clarify the goal of optimizing the CI critical path by adjusting the placement of a command. The AI responds well to this format, allowing for iterative refinement of the plan. By adopting this habit, the author notes that they are more likely to catch errors and inconsistencies that the AI might miss if they were passive. They mention that the single-prompt chat interface is limiting and encourages skimming rather than thorough review. Overall, the inline feedback method not only enhances the quality of the plans produced but also reinforces the author’s engagement in the development process.