No, most white-collar work is not "fully automated" by next Christmas

Dom · 18 July 2026 · Manchester

You've seen this claim, or one of its thousand children. Microsoft's AI chief told the Financial Times that most white-collar tasks — lawyering, accounting, project management, marketing — will be fully automated within 12 to 18 months. It has been screenshotted, quote-posted and threaded into oblivion, usually by accounts that end with a link to a course on surviving the apocalypse they've just announced.

Here's my problem with it. Not that AI isn't changing work — it plainly is, and I'll get to the uncomfortable numbers. My problem is the word fully, and the fact that the strongest evidence against the claim comes from the company that bet hardest on it.

Exhibit A: the company that went all in

Earlier this year Meta laid off around 8,000 people — roughly 10% of its corporate workforce — and reassigned another 7,000 into AI groups, including one with the genuinely chilling name of Agent Transformation. If anyone was going to prove that agents can absorb white-collar work at scale, it was a trillion-dollar company that restructured itself around the premise.

Then, on 2 July, Zuckerberg told staff at an internal town hall that AI agent development hadn't accelerated the way executives expected. That's not me saying it, and it's not a Luddite saying it. It's the man who did the layoffs, telling the survivors that the thing the layoffs were premised on is running behind schedule.

The honest numbers are uncomfortable — just not apocalyptic

Now the part the doomers get half right, because they always get it half right. Global tech layoffs passed 115,000 by May, nearly matching all of last year, with Meta, Amazon and Snap all citing AI as a driver. Entry-level tech hiring at top firms is reportedly down 30–50% against 2023. India's biggest IT firms are openly saying they'll never again hire at the volumes that built the industry. That's all real, and if you're trying to get your first job in tech right now, it's brutal.

But look at the shape of it. Brookings puts about 37 million US workers in the top quartile of AI exposure — and finds that only about 6 million of them face both high exposure and low capacity to adapt. The damage is concentrated in roles built around routine execution: the report-pulling, the first drafts, the account notes. That is tasks being automated. It is not professions being deleted. Those are different sentences, and the entire grift economy of X depends on you not noticing the difference.

One more number, because it's my favourite. A Udacity survey — 70% of respondents in management — found just 9% would want to replace their entire workforce with AI. The people who'd actually sign that purchase order don't want to. Turns out managers know something about their own jobs that thread guys don't.

What twelve years of service delivery teaches you

I've spent twelve-plus years in IT service management, which is the industry of finding out what work actually consists of once the PowerPoint version collides with reality. And here's the thing about knowledge work that every "fully automated" prediction misses: it isn't a pile of tasks. It's a chain of accountability.

An agent can draft the incident summary. It cannot own the major incident bridge at 2am when three suppliers are blaming each other and someone has to decide whose change gets rolled back. An agent can write the first pass of the root cause analysis. It cannot stand in front of a director and defend it. The automatable bit is real — it might genuinely be 40% of your week, and losing it will restructure teams and thin out entry-level rungs, which is the actual crisis worth talking about. But the accountability bit is the job, and nobody selling a course has explained how a language model carries the can.

So what do you actually do?

Not "learn to prompt" — that's the new "learn to code", half right and badly argued. The move, if you work in ops or service delivery, is to get yourself onto the supervision side of the line: workflow design, deciding what agents are allowed to touch in production, being the person who owns the outcome when the automation gets it wrong. The task-doers are exposed. The people accountable for the task-doers — human or otherwise — are not.

AI is coming for the boring 40% of your week. Let it. Just don't let someone with a course to sell convince you that the other 60% was never worth anything.

← Back to the chain