AI Doesn't Fix What You Can't Already Do

I brought in a designer to build prototypes fast, using AI. And on that front, he delivered. He generates screens quickly, asks the right technical questions, moves through iterations without friction. If I were only grading his AI skills, no complaints.

But none of that touched the actual problem.

He still doesn't communicate. I asked for step-by-step, incremental check-ins — show me direction early, before going deep on anything. Instead: silence. He disappears, doesn't answer, goes quiet for days. Then he resurfaces with a "final" version.

The final version has nothing to do with the brief — just something he personally felt like building.

My first instinct was to blame the speed. Like AI moving fast made him careless, made him skip steps he'd otherwise have taken. But that's not what happened, and it's worth being precise about why.

AI didn't cause the disappearing. AI didn't cause the ignoring of the brief. It just made the wrong thing faster and better-looking to produce — which makes the miss more expensive.

Take away the AI tools and this designer still doesn't check in, still doesn't work in increments, still builds whatever he personally wants to build. That behavior predates the tool. AI didn't introduce it. AI just removed the friction that used to slow it down — the friction that used to buy me a few extra days before I found out the direction was wrong.

This is the part people get backwards when they talk about AI and teams. They treat it as an equalizer — as if giving everyone the same tool makes everyone's output converge toward the same quality. It doesn't. It's a multiplier, and multipliers don't equalize, they widen.

Someone who's disciplined, who communicates, who respects a brief — AI compounds that. They ship more, in the right direction, faster than before. Someone who doesn't do those things gets the same multiplier applied to the opposite outcome: more output, produced faster, aimed at nothing you asked for, delivered with more confidence because it looks more finished.

The tool doesn't know the difference. It amplifies whatever you feed it.

So I've changed what I actually screen for. I used to ask candidates and collaborators how fast they are with AI tools — the wrong question, or at least an incomplete one. Now I ask what they were like before the tools existed: did they communicate on their own, did they work in checkpoints, did they respect scope when nothing was making that easy for them. That answer doesn't change when you hand them a better tool. It just gets louder.