AI is coming for the admin desk first. That’s where people learned.

Part of Measuring behaviour change.

The corner office can wait

The usual picture of AI at work is a threat to the top: the strategist, the analyst, the expert whose judgement suddenly looks automatable. It makes for a good headline. It isn't where the change is showing up first.

AI isn’t coming for the corner office first. It’s coming for the admin desk.

Think about what happens at that desk. Someone drafts the minutes, chases the missing figures, formats the report, reconciles two versions of the same spreadsheet, writes the first pass of a reply to a difficult customer. None of it is glamorous. Most of it is exactly the kind of structured, repeatable work that AI tools now handle quickly and tidily.

What the ONS figures say

The Office for National Statistics asked UK businesses with ten or more employees that reported using AI technologies in September 2026 which roles were most affected. Thirty-eight per cent said administrative and clerical roles [1].

The second figure matters just as much. Only six per cent reported reduced headcount as a result [1].

Put those two numbers side by side and the story is less dramatic than the headlines. In these figures, AI is mostly changing jobs, not cutting them. What is moving, mostly, is what people spend their day doing.

That sounds like good news, and in one sense it is. But it raises a question that a headcount figure can't answer: what happens to the learning that used to come with the work?

The desk was the classroom

Those admin and junior tasks were where people learned.

Drafting the minutes taught you how decisions actually get made, and who makes them. Chasing the figures taught you where the data lives and why it's usually late. Writing the first draft of a tricky reply, and having it torn apart by someone more experienced, taught you tone, judgement and the organisation's unwritten rules. Nobody called it training. It was simply how capability formed: do the work, get it wrong, get feedback, do it again.

When AI takes the first draft, the junior person's role often shifts from producing the work to checking it. On paper, that looks like a promotion. In practice, it can remove the very step where expertise was being built.

Approving is not producing

Johnson Wong put it precisely in Chief Learning Officer: "Approving work is not cognitively equivalent to producing it." [2]

Reviewing a well-formatted draft feels like engagement. You read it, you spot the odd error, you sign it off. But you haven't had to structure the argument, choose what to leave out, or wrestle with the blank page. Those are the struggles where judgement develops.

Hand that to AI, and future experts skip the part where expertise forms.

This isn't an argument against using AI for admin work. The efficiency is real, and nobody needs to format reports by hand just to keep the practice. It's an argument for noticing what else that work was doing, and making a deliberate choice about it rather than losing it by accident.

Where practice and feedback still need to happen

So decide, deliberately, where people still need practice and feedback.

For most L&D teams, that starts with a few practical questions:

• Which tasks used to teach people the job? List the junior and admin tasks in one team, and mark the ones that built judgement, not just output.
• Which of those has AI now taken over? Be honest about where people have moved from producing to approving.
• Where must someone still produce the work themselves? That might be one draft a week, one report a month, or a regular exercise without the tool, followed by feedback from someone experienced.
• Who gives the feedback, and when? Practice without feedback is just repetition. Name the person and the moment.
• How will you know if it's working? Look for what people can do unaided after a few months, not how quickly they sign off AI drafts.

None of this needs a new programme. It needs someone to look at how work is changing and ask what learning went with it.

Measuring how capability builds

This is the problem the Learning Velocity Index™ is built to look at. It puts Competence and Resilience in one score. It's a self-assessment of how fast and sustainably an organisation builds capability, which is exactly what is at stake when the tasks that used to build it are changing.

Like any self-assessment, it's a first read rather than a final verdict. If you want a starting point for the conversation, get your number free — the self-assessment link is on this page.

What this means for your programmes

• Treat "AI is changing jobs, not cutting them" as the start of a learning question, not the end of a workforce one.
• Map where junior and admin tasks used to build capability, and which of those AI now handles.
• Protect deliberate practice and feedback in the places where expertise still needs to form.
• Remember that approving work is not the same as producing it.

Figures from the public sources linked. No affiliation with or endorsement by the organisations named.

General information on training design, not professional advice for your organisation.