What to Keep Doing by Hand

Skills decay when they go unused. That isn't a claim about AI — it's how expertise works, and it's why surgeons who stop operating get slower and translators who stop translating get rusty.

What's new is the speed at which AI lets you stop using several skills at once, without ever deciding to. Nobody chooses to get worse at first drafts. They accept a good-enough draft on a busy Tuesday, then again, and eleven months later the blank page is harder than it used to be.

That may be an entirely acceptable trade. Plenty of skills are worth losing — nobody mourns their mental arithmetic. The problem isn't the trade. It's that it happens by drift rather than by decision, so you find out which skills you gave up at the moment you need one back.

The distinction that matters

Not all delegated work costs you the same thing.

Some skills are pure output. Formatting, boilerplate, converting between formats, first-pass structure. Handing these over costs you nothing you'll miss. If you never write another meeting agenda from scratch, no capability erodes.

Some skills are how you think. Drafting is the obvious one — writing is how most people work out what they actually believe, which is why a draft you didn't write is a position you haven't taken. Analysis is another: the value isn't the chart, it's what you noticed while making it.

The second category is where atrophy costs something real, and it's harder to see, because the output looks the same either way. A document produced by someone who thought it through and one produced by someone who accepted a draft are indistinguishable on the page. The difference shows up when someone asks a question the document doesn't answer.

The test

For any task you've handed over, ask: if the tool disappeared tomorrow, could I still do this to the standard my job requires?

Three possible answers.

Yes, easily. Delegate freely. You've kept the skill; you're just not spending time on it.

Yes, but slowly. Fine, as long as you know. This is the largest category and it's stable — you'd be rusty, not incapable.

No, and I used to be able to. This is the one worth acting on. Either you decide the skill genuinely doesn't matter any more, which is a legitimate answer, or you keep a deliberate practice going.

The point of asking isn't to use AI less. It's to know which answer applies before something forces the question.

Keeping a skill without giving up the speed

You don't have to choose between hand-writing everything and losing the capability. A few things work:

Do it cold first, then compare. Write your version, then ask AI for its version, then look at the difference. You keep the muscle and you get the speed benefit, at the cost of a few extra minutes on work where it matters.

Keep one of each kind. One document a week you write yourself. One analysis you do by hand. Not everything — a representative sample of each skill you've decided is worth keeping.

Use it as a tutor rather than an answer machine. Asking "what am I missing here?" keeps you in the problem. Asking "write this for me" takes you out of it. The output is similar; what happens in your head isn't.

Sofia — the skill she'd stopped noticing

Sofia is a senior policy analyst at a research institute. She'd been using AI for first drafts of briefing notes for eight months, editing them heavily, and considered the arrangement a clear win.

What surfaced it was a live session with a minister's office where she was asked to work through an argument on a whiteboard. She could do it — but she noticed she was reaching for structures the AI habitually produced rather than the ones she'd have chosen herself. Her thinking had absorbed the shape of the tool.

She hasn't stopped using it. She now writes the argument's spine — the three claims and their order — by hand before opening anything, and lets AI do the prose. Her estimate is that it costs her fifteen minutes and returns the part she'd been quietly losing.

Idris — the decision he actually made

Idris runs marketing at a 60-person software company. He audited what he'd delegated and found he'd stopped writing campaign copy entirely.

He decided that was fine. Copy wasn't the skill his job was judged on — positioning was, and he was doing more of that than before. Writing it down turned a drift into a choice, and it changed what he asked his team to protect: they keep the positioning work, and hand over the production.

The audit's value wasn't in reversing anything. It was in knowing which of his capabilities were load-bearing.

The one thing

Losing a skill to AI is fine. Losing one without noticing isn't, because you find out at the moment you need it.

Ask the question once a quarter: could I still do this? Then decide. Both answers are legitimate — drift isn't an answer at all.