How I've Been Using AI - Accidentally in Code

The core idea is that AI works best as a focused productivity assistant for leaders who already have a clear point of view. The author writes drafts, then asks the model to refine language and tighten arguments, limiting the session with a "stop when diminishing returns" cue. This habit lets fast writers spend less time on polishing and more on thinking.

A second practical pattern is using AI to generate and improve feedback on pull-requests or documents. By feeding the initial thoughts and asking the model for additional angles, the leader can read a piece once and get a richer critique, saving repeated reads. The same trick applies to tone-adjustment: the leader tells the model to adopt a "busy engineering leader" voice and then soften it, offloading overthinking about how harsh the message sounds.

When the author tries to use AI for planning or structuring unfamiliar work, the results are mixed. Generating a book promotion plan produced an overwhelming to-do list that had to be trimmed, and a proposal outline felt shallow until the author iterated with the model. The lesson is that AI can jump-start structure but still needs human expertise to prune and deepen the output.

Overall the piece warns that AI is a productivity boost for known domains but can create an illusion of knowledge in unknown areas, leading to wasted effort or bad decisions. Leaders must treat AI output as a draft, validate facts, and avoid letting the model dictate direction.

The takeaway for technical leaders is to adopt AI as a disciplined editor and feedback partner, not as a generator of content or strategy. Use it to cut friction, reduce overthinking, and keep communication clear, while always applying a human filter for accuracy and relevance.