Taking it slow with AI

I've been taking it deliberately slow with AI tools. I was slow to adopt them and since doing so I've been taking my time introducing new concepts and features.

The main driver of this has been fear of skill atrophy. I'm a big believer in repetition to drill new learnings and cement knowledge (searing hot take) which means that handing off work to an AI agent seems like robbing myself of a chance to practise. I'm trying not to be too rigid with it, but on the balance I'm still keeping quite a tight grip on the "doing" side of the work and using it more for research. I expect this grip will probably loosen over time.

I'm reading a lot about how others are using these tools in their day-to-day. The idea of various subagents and personas performing different tasks and posting PRs while you go for a walk is undeniably cool and futuristic (and leaves me with a bit of FOMO), but I'm still a bit wary of a scenario where I'm not in the loop for the majority of the work. My limited experience with this approach makes me think I would begin to drift away from understanding the larger systems that I work with, as the introduction of new code outpaces my ability to meaningfully engage with it.

I suppose this comes down to slow deliberate adoption rather than some self-imposed dogma of how I should only be using these tools in a limited, "pure" way. Taking it slow means really soaking up each step of the way and considering what each new feature and technique means for my overall workflow and how I want to be approaching my work.