Three Modes I Want to Use AI In

Part three of a three-part series on AI and craft — after AI Dependence and the Deskilling of Developers and Losing the Hands-On Feeling.

There are three modes I want to use AI in during development:

Mode 1: AI does all the work / is a code monkey

• I want Claude Code to handle the planning, development, testing, PR raising, etc. Everything.
• I want to only write a good Linear issue/ticket description with feature scope and definition-of-done lists.
• I want the feature/tasks to just work once done.
• I want the AI to fix its own mistakes and loop until everything works.
• I want a supervisor model to spawn its own task-specific agents as necessary.
• This is for work I find tedious or don't want to do myself.
• This is also for quick prototyping of ideas where the end product matters more than my understanding of its workings, or when the logic is simple enough that I don't want to spend the time to develop/type it myself. I don't want or need to learn anything, be it because of lack of interest or lack of time.

Mode 2: AI is a true pair programmer/helper

• I want AI to hover over my shoulder while I do all or most of the work.
• I DON'T want auto-complete style suggestions in my IDE where it tries to auto-complete my coding thoughts. When this happens it feels like I am not the one coding, and it doesn't become part of my learning brain.
• I want its input only after my own — be it during brainstorming, planning, requirements specification or development.
• I want to be able to ask, "what do you think?", get its feedback, and choose whether to act on it or not.
• I want to be able to implement any suggestions myself, type it out myself so it becomes part of my learning brain.

Mode 3: AI is an instructor

• I want AI to hand-hold me through new domains, let me build up my own first principles.
• I want it to act like a tutor/instructor/teacher: tell me just enough to get going but let me figure things out on my own.
• I want AI to test me during the process by assigning exercises and asking questions, so it can tell whether I learned properly or not.

An aside: becoming part of my learning brain means I need to be physically involved in the doing, so that it becomes intrinsic to me over time — so it feels like I did the work. It's the same feeling I have when I actually write things down or work things out on pen and paper: it becomes indelibly part of my experience/brain/psyche.