AI workforce automation

A job is a bundle of tasks. Automation removes tasks from the bundle, and the job changes shape rather than disappearing — unless the bundle happened to contain only one kind of task.

This is why confident predictions about which jobs vanish keep missing. The roles most exposed are not the ones that sound most automatable; they are the ones whose bundle is narrowest, which is a different and much less obvious list.

It also explains why the same job title can be heavily affected at one employer and barely touched at another. Two people with identical titles can have completely different bundles, and the bundle is what determines exposure.

The pattern is consistent across organisations. Tasks disappear when they are repetitive, have a checkable output, and carry a low cost of being occasionally wrong. Drafting routine correspondence, first-pass document review, summarising, categorising, extracting fields from unstructured text.

Tasks survive when being wrong is expensive, when accountability must sit with a person, or when the work is fundamentally about persuading or negotiating with someone. Not because a model cannot attempt them, but because someone has to be answerable for the outcome.

The clearest predictor is where the context lives. Work whose inputs are all written down somewhere is exposed; work that depends on knowing who to ask, what happened last quarter, or what the client actually meant is not, because the necessary information is not available to be automated.

• Automated: drafting, summarising, classifying, extracting, first-pass review
• Retained: final judgement, accountability, negotiation, relationships
• Retained: anything where an error is expensive to discover late

The recurring surprise is that automating a task often increases the work rather than reducing the workforce. When drafting becomes cheap, more drafts get produced, and someone has to review them.

Review is also slower and more senior work than production was. Organisations frequently end up with the same number of people doing a different and more demanding job, which is a real change and not the one that was projected.

There is a well-documented historical version of this. Cash machines were expected to eliminate bank tellers; branches became cheaper to run, so more branches opened, and teller numbers held up for years while the role shifted towards sales and advice. The task went; the job re-formed.

The routine middle of a job was rarely only tedious. It was also how people learned the domain — the first drafts a junior wrote badly and had corrected, the simple tickets that taught how the system fit together, the document review that built pattern recognition.

Removing it improves this quarter’s throughput and removes the mechanism by which the next generation of senior people is produced. Organisations feel that in three years, not three months, and by then the cause is hard to attribute.

The employers handling it well are deliberately manufacturing what used to happen incidentally: reviewing generated output as a teaching exercise, rotating juniors through problems that have already been solved, and making the reasoning behind decisions explicit rather than assuming it is absorbed.

The useful exercise is to list your week honestly by task and mark which ones have a checkable output and a low cost of error. That is your exposure, stated concretely instead of as anxiety.

Then look at what remains. If the residue is substantial — judgement, ownership, relationships — your job is changing rather than ending. If the residue is thin, that is worth knowing early, while there is time to widen the bundle deliberately.

Widening the bundle is usually available without changing employer. Taking the part of the process nobody owns, getting closer to the customer, or becoming the person who can tell when an output is subtly wrong are all things that can be started this quarter.

Deployment is slower than capability, and always has been. Regulation, liability, integration with systems nobody wants to touch, and plain organisational inertia slow adoption far more than technical limits do.

That argues for neither complacency nor panic. The strategies that pay off — deepening judgement, owning outcomes, building relationships — are the same ones that pay off if none of this arrives on schedule, which is what makes them worth choosing.

The genuine risk is unevenness rather than the average. Aggregate employment holding up is little comfort if your specific role, employer or region is on the wrong side of the distribution, and that is the part worth planning around because it is the part you can act on.