Two years ago the question employers asked us was whether AI would replace roles. They have stopped asking it. The question now is narrower and harder: which parts of a job are still worth paying a person to do, and how do you prove someone can do them.
The tasks went first, not the jobs
Across the 4,100 role descriptions our employer councils reviewed this year, almost none disappeared outright. What changed was their composition. Drafting, summarising, first-pass analysis and boilerplate production have moved substantially to automated tools. What remains is the part that was always hardest to describe on a job spec — deciding what to do, judging whether the output is right, and carrying the consequence when it is not.
The work did not shrink. It concentrated. Everything easy to describe got automated, and what is left is the part that was never written down.
Four capabilities employers now screen for
- Framing — turning an ambiguous business problem into something a person or a system can actually work on.
- Verification — checking output against reality rather than against plausibility, which is a distinct and teachable skill.
- Accountability — owning a result produced with assistance, including when the assistance was wrong.
- Communication — explaining a decision to people who will not read the underlying work.
None of these are new. What is new is that they are now the majority of the role rather than the finishing touch on it, and that employers have started screening for them explicitly rather than assuming they arrive with seniority.
Why proof matters more than it did
When the routine layer of a job is automated, a CV describing that layer tells a hiring manager almost nothing. A candidate who lists the tools they have used is describing the part of the work that no longer differentiates anyone. This is the pressure that has pushed competency-based certification from a nice-to-have to something employers actively ask for.
It is also why the Bureau of International Professional Standards standards weight demonstrated work so heavily. An assessor does not ask whether you know what a technique is. They ask what you decided, why, and what happened — which is precisely the part an automated system cannot supply on your behalf.