AI accelerates expertise. It doesn't replace it.

A gardener who has just bought a mitre saw isn’t suddenly a joiner. And someone with no real experience, armed with a shiny new AI tool, isn’t suddenly a consultant — even when the output looks convincing enough to fool a few people.

I say this as someone who uses modern AI every day and rates it enormously. But it’s worth being precise about what it does, because boards are making real decisions on the back of it.

The real risk isn’t replacement

The fear usually voiced is that AI will replace battle-hardened experts. That’s not the risk I see. The risk is the opposite: non-experts confidently publishing polished nonsense they don’t understand, can’t validate, and can’t defend the moment someone asks a hard question.

A tool can make you faster. It can make you produce more. It can even make you look competent. What it can’t do is give you the judgement to know when it’s wrong.

A near-miss worth remembering

Not long ago an internal team was wrestling with serious problems on their core network directory — the system that quietly authenticates every user and device in the building. Under pressure, they’d asked an AI to triage it. As usual, it came back fast, structured, and supremely confident, with exact step-by-step instructions.

It was also wrong about something fundamental — the way the system keeps its copies in sync. Following that confident advice would have destroyed their only healthy copy and triggered a multi-day recovery. Except recovery wasn’t an option either: the AI never thought to check the backups first, and those had been silently failing for some time.

Credit to the team — they paused and ran it past someone who’d spent years fixing exactly this kind of mess before touching anything. That instinct is the whole story. The fix, once you’d seen it before, was straightforward. The danger was entirely in acting on confident output without the experience to spot the flaw.

What this means for leaders

AI is a superb accelerant in expert hands — it speeds up people who already know enough to catch it when it drifts. In inexpert hands, it manufactures confident mistakes at scale, faster than anyone can check them. The scarce ingredient isn’t the tool. It’s the judgement to know what not to do, and that still can’t be downloaded.

So adopt it — but govern it deliberately. Decide, as a leadership question rather than a tooling one, where confident-but-unverified output is acceptable and where it absolutely isn’t. Anything touching identity, production systems, money or recovery sits firmly in the second category, with a human who’s been in the trenches between the suggestion and the action.

Used well, AI makes good people better. It just can’t make people good.