Every time I read another headline about AI “taking jobs,” I want to ask the author one question: whose job, exactly? Because after 25+ years in technology leadership — financial services, insurance, government, B2B SaaS — I can tell you the dirty little secret that nobody wants to say in the boardroom.

Most of the work everyone assumes is “covered” isn’t.

Not because people aren’t good at their jobs. Because no team, anywhere, has ever had the capacity to do more than skim the surface. You cover the top suppliers. The biggest contracts. The loudest risks. The rest? You do your best with what you’ve got, and you hope.

I’ve been that SME. I’ve been the person management believed had it handled. And here’s the truth: it wasn’t covered. Not even close. It couldn’t be. Not at human capacity, not on human hours.

The Coverage That Never Gets Counted

Here’s the bit people push back on when I say this out loud: “But nothing ever went wrong. So it must have been fine.”

No. It wasn’t fine. It was absorbed.

This isn’t a one-off story. Every few years, across 25+ years, something has hit us — and it got absorbed the same way every time: a 72-hour stretch, a small team burning themselves out to stop it becoming a public incident, a close call that never made it past the four walls of the team because someone caught it at 2am instead of letting it land on a client’s desk. Different crisis, same pattern, on repeat across an entire career. Those stories don’t show up in a risk register. They show up in grey hair.

They’re supposed to show up somewhere, technically. There’s meant to be a near-miss register — a kind of pre-risk risk register, catching exactly this sort of thing before it ever becomes a real one. But in my career, these issues rarely see the light of day, let alone make it anywhere near a near-miss register. They get absorbed, not logged.

The gap wasn’t zero-risk. It was invisible-risk — quietly closed, over and over, by a handful of people burning themselves out as the backstop for a system that was never actually designed to hold.

Here’s what that actually looks like, stripped of names and details: a supplier renewal clause nobody flagged until three days before expiry, because the person tracking it was covering forty other contracts and this one wasn’t “top ten” by spend. A late night spent reverse-engineering what should have been caught weeks earlier, because the alternative was explaining to a client why it wasn’t. Multiply that by every team, every sector, every year — and you start to see how much of “business as usual” is actually just very good improvisation.

And the grey hair is the audit trail — one that gets a new entry every few years, like clockwork.

What AI Actually Replaces

This is why I think the “AI is coming for your job” framing misses the point entirely. AI isn’t replacing the work your team was doing. It’s doing the work that was never being done at all — the long tail of suppliers nobody had time to review, the contract clauses nobody had bandwidth to check twice, the obligations tracked in someone’s head instead of a system.

It’s not a threat to competent people. It’s a threat to the fiction that the skim-level was ever enough.

Now, the honest objection: “that’s exactly what they said before the redundancies.” I get it. More coverage sounds great until it’s used as the business case for a smaller team. And yes, we’ve already seen companies publicly pin redundancies on AI. My honest read: most of that is premature, and a fair chunk of it is cover for a downsizing decision that was coming anyway for entirely different reasons. It’s far too early to claim you can systematically bank that much headcount saving from AI alone.

What we’re actually seeing right now isn’t role-based erosion. It’s task-based erosion — and that’s a very different thing. Tasks that used to be done entirely by human hands are getting augmented and accelerated. That’s a good thing. Nobody’s whole role is disappearing; the individual tasks inside it are changing shape.

Which means the real work for leadership isn’t deciding who to cut. It’s rethinking how roles should be formed and scoped now that AI helpers and agents sit inside the team. In our own business, we’re collapsing product management and QA testing into a single role — not because we don’t value either discipline, but because AI accelerates both the design and definition stage of product work and the acceleration of QA testing itself, so the same person can now credibly own both ends of that pipeline. That’s role redesign driven by capability, not headcount cut driven by fear.

That’s the distinction I think gets lost: task acceleration is happening now, at pace. Role elimination, at scale, off the back of AI alone — we’re not there, and anyone claiming otherwise is telling you more about their spreadsheet than about the technology.

What a True SME Already Knows

Here’s the tell. If you’re a genuine subject matter expert — not just holding the title, but someone who’s actually lived the skim-level, felt the 2am close calls — you recognise everything I’ve just described immediately. You don’t get defensive about it. You welcome the help.

The people who get threatened by AI covering the gap are usually the ones who were relying on the gap not being noticed. The people who’ve actually done the job know exactly how thin the coverage was, and they’re relieved someone — something — is finally watching the parts they never had the hours for.

But let’s be straight about who AI actually is a threat to: people coasting through their role. Nine to five, no drive, no push toward genuine subject matter expertise, no interest in using AI to get more out of their own capability. If you’re not trying to add to your role, stretch what you can cover, or become better at what you do — AI is going to expose that, and honestly, that’s on you. Not on the technology. The gap you were quietly relying on is about to get very visible, and that’s a fair consequence of coasting, not an injustice done to you.

Scott Galloway put a version of this well: AI itself won’t take your job — the person next to you who’s learned to use it properly will. That’s not a new law of nature. It’s the same law that’s always governed who gets ahead: the more skilled and more capable displace the less skilled and less capable. AI hasn’t changed that rule. It’s just raised the bar on what “capable” means.

So if the fear isn’t “will AI take my job,” the real question becomes: how can it help me?

That’s not a rhetorical flourish. That’s the actual next question — and it’s a bigger one than this post can answer. That’s blog two.

To be clear, none of this means AI covers everything. It doesn’t make the judgment call on which risk actually matters to your business. It doesn’t own the relationship with the supplier, or take the accountability when a decision goes wrong. That’s still you. What it does is make sure the decision you’re making is based on the whole picture instead of the ten percent you had hours to look at. That’s a different job than the one being described in the headlines.

There’s a real risk here too, and it’s the same one we’re already seeing play out in “vibe coding” — people shipping software they don’t actually understand because the AI wrote it. The SME still has to exist. Someone in that role needs to genuinely understand how the work gets done manually, without AI, so they know whether what the AI is producing is actually right — and so they can get the best out of it instead of just accepting whatever it hands back.

AI alone is a powerful force — it exposes those who lack real experience. SME alone is skilled and resilient — it can withstand pressure, just slower and harder-won. SME + AI together: stronger, faster, smarter, and built to last — deep foundational knowledge and lived experience, accelerated.

The risk isn’t AI doing the work. It’s putting an untrained, unskilled person in the seat who relies entirely on AI because they couldn’t do the underlying task themselves, even slowly. That’s when things go wrong — not because the AI failed, but because there was nobody capable of catching it when it did.

What AI actually gives you is scale: no fatigue, no nine-to-five, capability running twenty-four seven, one skilled person overseeing multiple AI agents instead of doing everything themselves. That’s a genuine productivity gain. But the overseer role only works if the person in it has a solid, grounded understanding of what those agents are actually doing — and, if push came to shove, could do it themselves, just slower. Take that grounding away and you haven’t scaled expertise. You’ve just scaled risk.

Because I know one thing for certain: given the choice between the surface-level coverage that cost me my hairline, and a system that watches the parts nobody had time for — I know which team I’d want to be on.

A Small, Slightly Embarrassing Confession

Funny enough, this post is a live example of the exact thing I’m describing.

I had this thought at a party. Ranting, a bit too passionately, to whoever would listen. In the old world, that’s exactly where it would have died — a fleeting idea, gone by the next morning, lost to the noise of a good night out. Nobody’s fault. That’s just how ideas usually go. You don’t get around to it, the moment passes, and the gap between “I had a thought” and “I did something with it” never closes.

Instead, I talked it through with AI in fragments over a few days — half-formed, correcting myself, adding the bits I’d missed — and it helped me turn a fleeting idea into something worth publishing.

That’s not AI replacing my thinking. That’s AI covering the gap between having a good idea and actually finishing it. The gap that, before now, nothing else ever covered.

And before anyone says “well that’s not really your work” — recognised experts have been doing a version of this for years. Jamie Oliver, a chef I’ve admired for a long time, is dyslexic and records his recipes instead of writing them; someone else helps turn that into the finished book. That’s closer to dictation than what happened here, but the principle’s the same: the ideas were always his, the format just needed help. What I did was a step further — a back-and-forth, AI pushing back on weak points, me correcting the record when it got something wrong. Less a transcription, more a sparring partner. Either way, countless SMEs and celebrities have leaned on someone else to help finish the work. Nobody calls those books fake. Nobody says the ideas weren’t theirs.

The only difference with AI is access. A ghostwriter costs money most people don’t have. AI just gave that same capability — turning a voice and an idea into something finished — to everyone else.

For what it’s worth, I went back and roughly tagged this piece line by line — which parts were my original thinking versus what AI actually added. By my own rough count, somewhere around two-thirds of the substance — the arguments, the examples, the claims worth disagreeing with — was mine, said in fragments over several days and at least one party.

I didn’t just take my own word for it either. I had two separate AI models — including one from a different company — independently audit the same question, without seeing each other’s answers. Both landed in the same place: roughly two-thirds of the substance was mine, and AI’s contribution was concentrated in a narrower place — sharpening arguments, adding a few extending ideas, and doing the work of turning ten scattered, out-of-order rants into one argument that builds instead of repeats. Neither audit let me off easy, either — both pushed back on places where I’d been a bit too modest about what AI actually added, and both flagged that a couple of the specific examples in this piece are illustrative rather than literal, which is a fair challenge worth being upfront about.

That’s the actual finding, checked twice, from outside my own head: the ideas were never the bottleneck. Finishing them was.

So here’s where I land: stop asking whether AI is coming for your job, and start asking whether your job was ever fully covered in the first place. If it wasn’t — and for most of us, it wasn’t — that’s not a threat. That’s the gap finally getting the attention it always needed, and never had the hours for.