For the last few years I've been the annoying friend. Every time I caught up with friends who work in tech, I'd bring up AI tools, because it was obvious to me that our field was changing and I just loved talking about it.

A lot of them weren't interested. The lines I heard most could be summed up as "I tried it and it was crap."

Honestly, they probably weren't wrong about what they tried. Usually it was a free model, or a paid one that was already a few months old, which in AI time is a lifetime ago.

The trouble is that the tools kept changing and the opinion didn't. Plenty of people I know have a paid ChatGPT or Claude account that's still running old models that have received massive upgrades months ago. I've seen this a crazy percentage of times to be honest. Nobody told them there's a dropdown.

Then some of them got laid off.

Watching friends get hit by the bus

Now they're job hunting, and the market has changed a lot in the last couple of years to the point it's almost unrecognizable. A lot of interviews now assume you work with these tools and being proficient is part of the screening. Skills that were plenty two years ago are behind, and they're finding that out at the worst possible time.

I don't feel any "I told you so" about this. (I really don't.) It sucks. These are good people. They had years to pick this up at their own pace, and the timing they ended up with is the hardest one.

That's why I'm writing this to the people who still have that time.

Every incentive at work says wait

If you're not using these tools much, it probably doesn't feel like a problem right now. For people who've been at their jobs more than a year or two, it mostly isn't one yet. And almost everything around you is telling you to wait.

Look at the incentives. Finish a two-week project in three days and your reward is a three-day deadline next time. If you bill hourly, it's worse: get four times faster and you've given yourself a 75 percent pay cut. Automate the tasks that fill your week, and there may not be another project waiting on the other side. Some of the people worried about that are right.

So people do the sensible thing. They keep quiet, keep working the old way, and the project ships right on schedule.

Teams hide the gap too. On a lot of teams, a few people lean on AI heavily and carry a big share of the output. Projects ship. The numbers look fine. Everyone's review says "meets expectations." You can go a long time without anything at work telling you the ground has moved.

None of that is a personal failing. These are real problems, and leaders should fix them.

But every one of those incentives belongs to the job you have now and not the macro trends. A new employer is hiring you, and they'll look at what you can do on your own and compare it with people who use these tools every day. That is the bus I watched some friends get hit by. It's also the one part of all this you actually control.

If you tried it and it still kind of sucks

Some of you are past the skeptic phase. You've been trying, and the results are underwhelming. It writes confident nonsense. Fixing its output takes longer than writing the thing yourself. You're starting to wonder whether everyone else is exaggerating.

That's the normal middle part. Almost everyone goes through it.

But nobody looks at a nice table and says the power drill built it. Using AI well is the same kind of skill. A beginner gets a lot of generic, subtly broken, very confident output. Someone with more hours gets something different out of the same model, because they've learned what context it needs, when to throw out its approach, and when the thing technically works and still sucks.

Those hours are most of the difference. When someone on your team builds something fast with AI, it's easy to assume the tool did it. Usually they've just put in more reps and settled on a workflow. Which means you can too.

Why I care how fast you get there

Selfishly, this matters to me too, because the gap shows up in how we all work together.

A lot of building software is one loop. You start with a problem, make some design calls, build it, use it, realize the idea was wrong, and go around again. AI made that loop much faster. It used to take half a day or more to prove an alternative to a design direction, so we argued before we built, because building was expensive.

Now I can sometimes get through three versions before the meeting about the first one wraps up. (This makes me extremely fun in meetings.)

I got quiet, and I shouldn't have

For a while I pushed hard on this. I'd prototype things we were still discussing and build alternatives nobody asked for. That creates friction with teammates, and friction is tiring, so eventually I stopped pushing so hard.

Once I slowed down, the old way looked like it worked fine, because nobody was showing what else was possible. That's part of why I'm writing this. Staying quiet felt polite, but it wasn't kind, especially after watching friends find out the hard way.

Open the dropdown

So here's my ask, as a friend. If you tried it once and wrote it off, give it another real shot. Pick the best model your account has, which might mean opening a menu you've never opened. Use it on actual work for a couple of weeks, not a quick test. If you've been trying and it still feels clumsy, keep going a little longer, and ask whoever on your team seems fastest to show you how they work. Most of us are happy to. I definitely am.

If it still isn't useful after that, fair enough. I'll stop bugging you about it. But an opinion formed on an older model a year ago is about a version that doesn't exist anymore.

And please do it while you still have a job. Learning this on a regular income, at your own pace, with people around to help, is so much easier than learning it between interviews. I've watched friends do it the hard way. I'd really like you not to.