I'm consulting for a team of about 20 developers right now. Most of them I wouldn't want on my own team.

And everyone above them believes AI will save the team. Give them Copilot, give them Claude, and suddenly everything ships faster.

That's not what I see happening. AI is a multiplier. It multiplies what's already there.

If someone is skilled, reliable and fast, AI takes them to another level. If someone is average, needs hand-holding, and ships features that come back three times for fixes, AI won't fix that. The broken thing just arrives faster.

And what I actually see with these people is something worse. They use AI to make their own work easier, not bigger. Same output, finished sooner, call it a day. So the company now pays for the AI on top of the salary and gets exactly what it was getting before. The multiplier works, it just multiplies their comfort instead of their output.

Code output isn't even the metric that matters. How many meetings do you need before they start working on something? How many times does the feature come back broken? How much micromanagement does the person need? Those are the real metrics, and AI doesn't move them.

So my math is simple. I'd rather pay one person $15,000 a month, a real professional, than three people at $5,000 to $7,000 each. That one person with AI will outperform all three combined. The three average ones can't get there, no matter what tools you hand them.

This is an observation, not advice. But after watching a 20-person team adopt AI, and after talking with friends from other companies who see exactly the same thing, I'm convinced of one thing.

AI doesn't set the ceiling. You do. Your skills, your reliability, your work ethic. That's exactly what AI works with.

To be fair, there's a flip side. It's pretty naive to expect people on a fixed salary to be motivated to deliver twice the output. And whether we like it or not, working this way is mentally much more demanding. AI delivers the things, but the context switching never stops.

That's a different question though, for a different time and a different post.