The skill you stop practising is the one you needed to check the output

Checking is not a lighter version of doing

The assumption underneath most confident use of these tools is that a person can evaluate output without being able to produce it. Sometimes that holds. Frequently it does not, because recognising what is missing from a piece of work requires knowing what should have been there, and that knowledge is built by doing the work.

This is familiar from other fields. An editor who has never written cannot see the structural problem; a reviewer who has never built the thing accepts the design that will fail in a year. Evaluation is a skill in its own right, and it is largely a by-product of practice at production.

The consequence is a dependency people rarely map. Reviewing generated work adequately depends on competence acquired by not using generated work, and that stock is not replenished by reviewing.

The erosion is invisible from inside

A skill that is fading does not announce itself. Judgement feels the same while getting slower and less reliable, and the tasks that would have revealed the gap are the ones now being handed over. Someone who has not written an argument unaided for a year does not experience themselves as worse at arguments.

It also plays out at the level of a team. Where the routine work was how newer people learned, automating that work removes the training path while leaving the experienced reviewers in place. The immediate output looks the same, and the problem surfaces years later when the reviewers move on.

None of this is an argument against the tools, and it is an argument for noticing what the practice was for. Plenty of tedious work was also the mechanism by which competence accumulated.

Which skills matter, and which genuinely do not

Not everything lost is worth keeping. Manual arithmetic, remembering syntax and formatting citations by hand were never the point, and no serious person mourns the previous generation of those. Tools have always absorbed mechanical skills and the profession usually improves.

The distinction that seems to hold is between skills that are purely executional and skills that carry judgement inside them. Formatting a reference is executional. Deciding which reference supports a claim is not, and that judgement is exercised precisely while doing the work that could be delegated.

Anything where the doing is how you find out what you think belongs firmly in the second category. Writing an argument, working through an analysis, debugging something unfamiliar: the output is a by-product and the understanding is the actual product.

Practical ways to keep the stock up

The simplest is to keep a category of work you do unaided, chosen because it is where your judgement matters rather than because it is difficult. For a writer that might be anything with a byline; for a developer, the parts of a system that are hard to change later.

A second is to attempt before delegating on anything unfamiliar. Spend ten minutes forming your own view first, then compare. It slows you down slightly, it keeps the evaluation muscle working, and it produces better prompts as a side effect because you now know what you are asking for.

A third applies to anyone responsible for how others learn. Protect some proportion of the work where people acquire judgement, and be explicit that it is protected for that reason rather than because of efficiency, since nothing about it will look efficient on a quarterly view.

None of these require abstaining from the tools on the work where they help most. What they require is that the choice of which competences to keep is made deliberately rather than settled by whichever tasks happened to be convenient to hand over. Convenience is a poor selector, because the tasks that are easiest to delegate and the tasks that built the judgement are very often the same ones.

The honest uncertainty here

Every previous technology that absorbed a skill provoked the same warning, and the predictions were wrong at least as often as they were right. Calculators did not end numeracy in the way feared, though the shape of numeracy changed. Confident claims about permanent degradation should be treated with the same caution as confident claims about pure gain.

What can be said without speculation is narrower. Skills are maintained by use, delegation reduces use, and the effect of that on any particular capability is an empirical question that a few years of practice will not settle.

The reasonable position is to be deliberate about it rather than to arrive at whatever the defaults produce. Decide which competences you intend to keep, keep practising those, and hand over the rest without much regret.