I have a simple test for whether I’m actually using my AI tools well:
Is anything I did this week something I couldn’t have done before?
Not the same work, faster.
Something I could not have done.
If the honest answer is “no, I did the same work, just quicker,” then I’m not using the tool for what it’s for. I’m using it to defend the ceiling I was already standing under.
And that ceiling usually isn’t the tool’s. It’s mine — a list of things I decided long ago were too slow, too hard, or outside my skills, and stopped considering. The tool has changed since I made that list. The list hasn’t.
The comfortable use of a powerful tool
The easiest way to use an AI tool is to point it at work you already know how to do.
- Write the function I was going to write.
- Rename things.
- Generate the tests I would have written anyway.
- Translate this snippet to another language.
There’s nothing wrong with any of that. It’s real time saved, and time is real.
But it’s also the shallow end. Autocomplete got faster; the shape of my work didn’t change. And if the shape of my work never changes, I’ll never find out what the tool is actually for.
What “can’t” looks like
“Can’t” covers more than skill. Some of the work I’ve never done really was beyond me. Much more of it was blocked by cost — of time, of attention, of risk, of finding out.
AI lowered both. So the list worth looking at isn’t just “things I’m unable to do.” It’s:
- Things outside my skill — the language I don’t write, the part of the stack I’ve never touched.
- Things too slow to justify — the refactor across twenty files, the migration I’d have scheduled for “someday.”
- Things too broad to hold in my head — reading a whole unfamiliar codebase before touching it, comparing three approaches properly instead of picking the first one that compiles.
- Things too risky to try — the experiment I’d talk myself out of because a failed attempt would cost a week.
- Things too small to be worth my hands — the throwaway script, the one-off tool, the visualization I’d never have built.
- Things I never thought to ask for — the workflow I kept doing by hand, because automating it had always cost more than just doing it.
Every item there used to sit on the wrong side of “worth trying.” Speed alone changes little. What matters is when the tool moves something across that line — because it got cheap enough, or because the skill gap closed. That’s the difference between speed and reach.
The last item is the one that surprised me. I was coordinating several coding-agent sessions by hand. It only took about half an hour, but it was boring: reading through every session’s discussion and carrying findings from one project to another.
It never occurred to me to build a tool for it. A tool for a half-hour chore had
never been worth building. But I was already using coding agents all day; the
cost of building one had collapsed, and I hadn’t noticed. When I finally did,
it became /discuss.
The cost had dropped long before my sense of what was worth attempting caught up.
The trap on the other side
This cuts both ways. AI makes bad ideas cheap too, and “I couldn’t have done this before” is not a reason to do it.
Doing more is not the same as doing better. The judgment moves up a level: if the tool removes the cost of trying, then the scarce thing is no longer execution — it’s deciding what deserves to be tried. That’s a harder job, and it’s still mine. So is understanding what I ship — most of all when it’s work I couldn’t have done on my own.
So the test isn’t “did I attempt something ambitious?” It’s “did I attempt something ambitious that was worth attempting, and could I only attempt it because of the tool?”
The question to keep asking
Every so often, look back at what you shipped. If all of it would have been possible a year ago without the tool, you’ve bought speed, not reach.
The point of a tool that removes a constraint is not to do the same work in less time. It’s to spend the time you got back on the work the constraint was hiding. So go looking for it: pick one thing you decided long ago you couldn’t do, and try it this week.
If you’re only doing what you already could, you’re treating the tool’s floor as your ceiling.