I'm losing skills on purpose

Technical debt is a tool, not a failure. So is skill debt: the skills you let rot on purpose. The real risk is borrowing competence with no apprenticeship left to repay it.

I lean on AI as hard as anyone I know, and it is making me worse at things. That is not a confession; it is the same trade I made with GPS years ago: I can no longer hold a map of a new city in my head, and in exchange I can drive anywhere on earth without a moment of preparation. I made that trade on purpose, and I would make it again.

There is a loud argument online right now about whether AI is rotting our skills, and what that will cost us. It is a real worry, but it is not a new one. Every serious tool set off the same alarm: the calculator, the word processor, the search engine. We have been letting skills rot for a long time, and mostly it was fine. The useful question is which ones to carry forward, and which to drop without a second thought. We have always had a rough answer. It is the oldest map there is: the career ladder. You start by doing the work, you earn the right to check other people’s, and eventually you choose what gets built. Doing, checking, choosing. AI left the ladder intact, but made the bottom rung suddenly cheap and abundant. So the real question left is which skills to keep paying down, and which to let slide into debt.

Atrophy is a promotion

You let a calculator do your arithmetic, you get slower at sums, and it does not matter, because the sums were never the point. The calculator made arithmetic cheap, the spreadsheet made it free across thousands of numbers at once, and unlocked work nobody could do by hand. You got promoted to the work on top.

The same is happening to me at work. After a heavy year of leaning on AI, I have lost recall of technical details I used to keep in my head. These days I ask for the small implementation facts I used to keep loaded: the API shape, the syntax edge case, the command I know I have used before but no longer remember cold. I do not mourn much of that loss, because the understanding underneath still lets me validate the details when AI fetches them, fresher than memory ever kept them. And the slide started long before AI. In my last year in big tech I wrote almost nothing myself, because by then my job was deciding what the teams should build, not building it. That is what moving up has always cost.

But this is exactly where it gets dangerous, because some of what you drop is not dead weight. Some skills you let go and you get lighter and faster. Others you let go and you quietly hollow yourself out. Telling those two apart is the whole problem.

Engineers already have a word for this kind of deliberate decay. We call it technical debt: the shortcuts you take on purpose, everywhere the polish is not worth it, because gold-plating every corner is its own failure, slower and bloated and worse at the few things that actually matter. The software that runs the world is full of debt by design. The skill is choosing where to take it and where to pay it down. Skills are the same. Letting a skill rot is not failure. It is debt. The only question is whether you designed it or drifted into it.

Doing, checking, choosing

The ladder has three rungs. Doing is producing the thing. Checking is knowing whether the thing is any good. Choosing is deciding what should exist at all. AI is doing something different to each, and so should you: take the debt freely in the doing, carefully in the checking, never in the choosing.

Doing is the production: the first draft, the routine version of whatever your trade makes. Hand it over. On routine work the machine is often already good enough, and if you once learned the skill you can get it back fast. And the unfashionable truth: in most work the craftsmanship was always a means to an end, much as we loved it. Where the craft itself is the point, you keep it. That is the exception, not the rule.

Checking is knowing whether the work is actually good: catching the quiet mistake, doubting the confident answer, sensing that something is off before it ships. Everyone now agrees this is the skill to keep, which is true and also where most people stop thinking. What they miss is that checking is built out of doing. You learn to smell a bad answer by producing thousands of them yourself. Once you have built that nose, it lasts; I am rustier than I was at producing the work, and it does not matter, because I can still tell when the work is wrong. What you cannot do is check what you never learned to make. The immediate danger is smaller for experienced people; they can offload plenty before they go blind. The deeper danger is the ones coming up, who never produce enough to build the instinct in the first place.

Choosing is deciding what is worth doing in the first place: what to build, which problem to frame, which details actually matter and which are noise. This is where the scarcity moved, because the machine does not know what you should want. The word for it is taste: the sense of what is good and what matters that no average can hand you. These are the skills I now train on purpose, because daily work no longer trains them by default.

The most valuable thing I ever learned was judgment, not any technology. Cross enough different kinds of work and you stop seeing each field as new; you start seeing the same handful of things that decide whether anything actually works. That is what lets you walk into a room on a problem you have never touched, spend a little while ramping up, and then hold your own with people who have lived in it for years. You usually know far less than them. What you have is a filter: which details matter, and which are just people defending their favorite opinions. That judgment carries from one domain to the next; the production skills do not.

Writing is how I sharpen my thinking, and I mean the thinking, not the prose. A vague idea sits comfortably in your head until you put it in a sentence, where it has to hold together or fall apart. Building is the same: you find out what you actually understand only when you force the idea into something concrete. So use AI to build, aggressively, because that loop from idea to working thing and back is exactly where the understanding gets sharp, and the machine makes the loop faster. Hand over the production, the prose and the typing. Keep the thinking. The trap is passivity: letting the machine think for you.

Originality belongs here too. An LLM pulls everything toward the average, which is often useful and sometimes fatal, because whatever in your work depends on a view that is only yours, your own particular wrongness, the model will quietly sand flat.

The debt you can’t repay

The important change is collective. Most of the worry online is personal: if I let AI do my work, will I lose my edge? Probably not. If you already built the skill, you will likely be fine, the way I am fine. If anything the individual risk may run the other way: clinging to a skill the machine now does for free, and getting quietly out-competed by the people who dropped it and moved on. The bigger risk is one level up. AI rents you the first rung, the doing, so anyone can now operate a rung or two above their experience. Individually that is a gift, and the rational move is to take it: jump up, spend less of yourself producing the work and more on judging and choosing it.

That is where the bargain turns. Technical debt is only safe because you can repay it: refactor, rewrite, put in the work later. Skill debt had a repayment plan too, and it was the apprenticeship, closing the gap between what you can produce and what you can judge by doing the work for years until the judgment was yours. AI lets a whole generation borrow the output while quietly removing that repayment. It is a tragedy of the commons: each of us is right to jump up and offload the doing, while together we drain the shared stock of judgment that the doing used to refill, and no one is paid to refill it. You can already see the first bills: people shipping plausible work they cannot actually explain. We are betting the judgment shows up without the work that built it.

Maybe the machine closes the gap itself, and the apprenticeship turns out to have been replaceable. For the bounded kind of checking, it probably does, and that is mostly a matter of time. Choosing is harder, and not only because it is hard. Choosing is also about power: people fight to keep the right to decide, to keep calling the shots, long after the machine could call them instead. And it scales with scope. Ask for a table built a certain way and the machine fills in the details; ask it to build you a nuclear plant and a human still has to decide what that even means. The machine takes the low-ambiguity calls. The high-scope ones stay ours, for now, and maybe for reasons that have nothing to do with capability.

What it comes down to

The hard part is no longer learning every skill. It is choosing which losses to permit. Do not keep every skill; keep the ones that keep your judgment. Attention has always been finite, but for most of a career that barely cost you anything: things moved slowly enough that you could keep a skill for its own sake, or climb at your own pace, and pay almost nothing for it. That is over. Now every hour you spend staying sharp at something the machine does for free is an hour not spent leveling up, and the gap compounds. Standing still now means sliding toward the work the machine already does. The opportunity cost of holding on went up, which is exactly why what you keep has to become a deliberate choice.

There is one last reason to be ruthless. The judgment you keep is the thing AI never lets rest. It works that one muscle all day, sharp at nine and depleted by noon. So rest becomes part of the work, including the boredom new ideas grow out of. The failure mode is trying to keep everything, spending all day in motion while the value drains out of it. AI makes that option too expensive. I am letting syntax recall, boilerplate, and first-pass prose decay. I am not letting product taste, architectural judgment, or the ability to notice nonsense decay. That is the bargain.