← All writing
aiengineering

AI in the loop, not in the driver's seat

How I fold AI assistance into real engineering without giving up review, security, or taste.

AI-assisted development is neither magic nor a threat — it's a power tool. Power tools are wonderful right up until you respect them less than they deserve.

Human-in-the-loop, always

Every AI-suggested change goes through the same review I'd give a teammate's PR: does it match the architecture, is it secure, does it handle the edge cases? The model drafts; a human decides. That boundary is non-negotiable.

Fundamentals don't get cheaper

If anything, AI raises the value of fundamentals. SOLID, clean code, semantic HTML, accessibility — they're how you evaluate what the model gives you. You can't review code you don't understand.

Where it shines

  • Scaffolding the boring 80% so I can spend attention on the 20% that matters
  • Exploring an unfamiliar API quickly, then verifying against the docs
  • Catching the typo-class bug before it reaches review

Used this way, it's not about writing more code. It's about spending judgment where judgment counts.