Which Parts of Your Job Are Actually Changing

The public argument about AI and jobs is conducted in whole roles: will this job survive, will that one disappear. It's an energetic debate and it's asking the wrong question, because a job isn't a unit of work. It's a bundle of tasks that happened to be given to one person, usually for reasons of logistics rather than logic.

AI doesn't take the bundle. It redistributes it — and the useful question isn't whether your role survives, but which of the tasks inside it are moving, and what the remainder is worth.

Sort the bundle by what makes the task hard

The distinction that predicts almost everything: is the difficulty in producing the thing, or in knowing what to produce and whether it's right?

Production-hard tasks are the ones where you know exactly what good looks like and the work is getting there. Formatting a report. Writing a first draft from a clear brief. Reshaping data. Producing five variants. These move, and they move fast.

Judgement-hard tasks are the ones where the work is deciding. Which question is worth answering. Whether this number is misleading. What the stakeholder actually meant. Whether a technically-correct answer is the right one to give. These don't move, because they depend on context that isn't written down anywhere — often on conversations no tool was in.

Most professionals discover, doing this honestly, that they've been spending a majority of their time on production-hard tasks while being paid for judgement-hard ones. That's the redistribution, and it's usually good news.

The three things that stay yours

Across every role, three categories resist handing over.

Accountability. Someone has to own the outcome. AI can't be accountable — not for legal reasons but for structural ones: it has no stake, and no capacity to be wrong in a way that costs it anything. Where your job involves being the person who answers for a decision, that part is durable.

Relationships and unstated context. A great deal of what makes a professional effective is knowing that this stakeholder needs the risk framed first, that team is already stretched, and last quarter's number is distorted for a reason nobody wrote down. None of that is in any document.

Deciding what's worth doing. Prioritisation is the least automatable thing in most jobs and the least discussed. AI will do whatever you ask, thoroughly. It has no view on whether you should have asked.

What this means practically

If most of your time goes on production-hard work, your job is changing significantly, and the direction is toward the judgement-hard part of the same bundle. That's a real change and it's worth preparing for, because being good at production doesn't automatically make you good at judging.

If most of your time already goes on judgement, your job changes less than the discourse suggests — but the production-hard tasks around it get cheap, which means the expectations on volume rise.

Either way, the thing to develop is the same: the ability to evaluate quickly and correctly. That's what becomes scarce when producing gets cheap.

Ana — the audit that surprised her

Ana is a senior HR business partner at a 200-person firm. She'd been anxious about the jobs debate in a general, unfocused way.

She spent a week logging her time against the two categories. Production-hard: policy drafting, meeting summaries, standard correspondence, job description writing, first-pass CV screening — about 55% of her week. Judgement-hard: performance conversations, deciding whether a grievance had substance, reading whether a manager's account of a situation was complete, advising on a restructure — about 30%. The rest was scheduling.

Nothing in the second list was going anywhere. Almost everything in the first was. The anxiety didn't disappear, but it got specific, and specific anxiety is actionable in a way that general anxiety isn't.

Bram — the part he'd been treating as overhead

Bram is a project manager at an engineering consultancy. His audit came out the other way: most of his week was already judgement-hard — reading whether "probably fine" meant fine, deciding which dependency was the real risk, managing a client's expectations against a slipping date.

What he found was that he'd been treating the production-hard remainder — status reports, translating the same facts for four audiences — as unavoidable overhead rather than as delegable work. It was about eight hours a week.

He handed it over and used the time on the part he was actually paid for. His role didn't change. What he spent it on did.

The one thing

The question isn't whether your job survives. It's which tasks in it are production-hard, because those are the ones in motion.

Do the audit honestly and the answer is usually less dramatic and more useful than the debate suggests: a redistribution toward the part of the work you were being paid for anyway.