Anyone can prompt an AI into a working demo now — that's a baseline, not a skill. The real question isn't “can they build something,” it's “do they know what they built, and can they defend every decision in it.”
01
I write unit tests as a default, not an afterthought, and I stress-test systems under simulated load before I call them done. I've built things complex enough — a tree-walking interpreter, a distributed job queue with a stale-job reaper, an auth library that detects refresh-token theft — that untested code would've broken immediately and silently. It didn't, because I tested it before I trusted it, not after something broke.
02
Planning a build with an agent, deciding what it should own versus what I should own, catching when it's confidently wrong before that mistake ships — that's the actual skill now, separate from knowing how to code. Most people either refuse to use AI on principle or use it uncritically. I do neither. I use it deliberately, and I stay responsible for the result either way.
03
I wrote my first code on a phone, before I owned a laptop — no proper keyboard, no multiple monitors, none of the setup people assume you need to be productive. If the setup's not ideal, I adapt. I don't wait for ideal conditions to start doing good work, and I won't wait for them on your team either.