AI Isn't Smart or Dumb. The Person Using It Is.

I keep running into the same pattern, on more than one team, at more than one company, and it's worth writing down.

Someone starts leaning hard on AI. Every response, every update, every message goes through a model first. On paper that should make them faster and sharper. In practice, it's often the opposite: the output gets longer, more confident, and less trustworthy. Claims that don't quite hold up. Explanations that sound reasoned but fall apart the moment you ask a follow-up question. And whoever's accountable for that work ends up fielding the fallout.

It took me a while to name the actual problem, because "using AI too much" isn't quite it. Plenty of people use AI constantly and it makes them noticeably better. The difference isn't the tool. It's what the person does with it before it leaves their hands.

AI is a multiplier, not a substitute

This is the same lesson I wrote about with escalations and systems: AI doesn't fix a broken process, it just moves the same problems faster. The individual version of that is just as true. AI doesn't fix a person's lack of understanding, it just lets them produce more output while still not understanding. A long, confident, AI-written response isn't evidence of good judgment. Sometimes it's the opposite: judgment replaced with volume.

Give AI to someone who already checks their own work, and it makes them faster. Give it to someone who doesn't, and it makes the gap in their understanding bigger and harder to spot, because now it's wrapped in polished language instead of an obvious shortcut.

What using AI well actually looks like

The people who use AI well, on any team I've been part of, share a few habits, none of which are about the tool itself:

• They read what the model gives them before sending it, not after someone else flags it
• They can explain the reasoning in their own words, not just repeat the output
• They use AI to get to an answer faster, not to avoid forming one themselves
• Their responses get shorter with AI, not longer, because the model helps them cut to the point instead of padding around it

None of this is about prompting technique. It's ownership. The model isn't the one accountable when the answer is wrong. The person sending it is, and using AI well means never forgetting that.

Where this matters most: support

This shows up hardest in support and customer-facing teams, because the cost of a bad answer is immediate and visible. It's also where AI has the most obvious, genuinely useful role: as the first front foot, not the last word. Let it triage, draft, and handle the repetitive first-touch work at speed. That's exactly what it's good at. What it shouldn't do is become the final answer that goes out the door without a person actually understanding what's being said and standing behind it.

The teams that get this right treat AI as the fastest way to a first draft, not the fastest way to close a ticket. The teams that get it wrong end up with confident, AI-generated answers turning into a mess that someone senior has to clean up.

The takeaway

AI doesn't make people smart or dumb. It makes whatever they already are happen faster. Used well, it's a real multiplier on good judgment. Used carelessly, it's a chaos multiplier dressed up as productivity.