Spot on. If you genuinely wanted a more productive civil service, you'd start with the workforce, not the software. The problem was never a shortage of chatbots. Decades of flattened pay scales have hollowed out the senior grades of anyone with options, and that's the actual rot.
A capable SEO or grade 7 can earn half again as much walking across the road to a consultancy or a bank, so the people who can leave, do. What's left is a structure where talent is rationed and the gaps are filled by headcount: layers of people managing other people, sign-off chains five deep, meetings about meetings, work that exists only to feed other work. One properly paid, properly empowered specialist can do what a department currently spreads across ten posts, because most of those ten exist to coordinate, review, escalate and re-do what the other nine produced. That's the daily reality of a system that would rather hire two mediocre people on band rates than pay one good one what they're worth, because the former fits the grading structure and the latter needs a business case nobody wants to write.
Here's the bit they'll never put in the job advert. The quiet hope behind "AI-first" is that a chatbot will paper over the cognitive gap left by twenty years of failing to hire and keep clever people. They can't say it out loud, so they won't, but the logic is plain enough: if we can't afford the calibre of person who'd get it right first time, maybe the machine can drag the mediocre ones up to passable. Buy in intelligence by the licence seat because we priced the human kind out of the building. It won't work, because a model trained to produce plausible-sounding text is a remarkably good fit for an organisation whose real failure mode is already producing plausible-sounding text that falls apart on contact with reality.
There's a false economy buried in here too, and I'd bet the business case hasn't priced it. AI is being sold at a loss right now. The headline subscription figures sit nowhere near the true cost of running these models, and the providers are eating the difference to buy market share and lock customers in. The civil service, true to form, will mistake the introductory rate for the real one. By the time it's wired into every workflow, every department dependent, every process rebuilt around it, the discounts evaporate and the prices go where they were always heading. At which point you discover the running cost of "AI-first" comfortably exceeds the wage bill for the competent people you refused to hire, except now you're locked into a foreign vendor instead of paying British salaries with the money.
Classic public sector: take the cheap thing today, pay multiples for it later, act surprised every single time.
None of them can see the size of what they're walking into, because they're blinded by the bandwagon. Somewhere a focus group told them that "AI adoption" makes government look forward-looking and modern, and that was enough. The whole thing is vanity dressed as strategy. They want the press release about being a "global leader in AI government transformation" far more than they want a functioning civil service, and they'll wire a foreign-controlled, loss-leading, hallucination-prone technology into the load-bearing machinery of the state to get it. When that machinery starts producing confident nonsense at scale, in tax, benefits, immigration, anywhere a wrong answer ruins someone's year, it won't be a press release. It'll be a public inquiry. But that's three reshuffles away, and the photo op is now.
It compounds. Underpay the centre and you end up outsourcing the actual thinking to contractors and the big consultancies at three to five times the day rate you refused to pay an employee. You spend far more for worse institutional memory, because the expertise walks out of the building every time the engagement ends. The Treasury knows this perfectly well, which is why the same departments rehire the same firms to redo the same "transformation" every few years.
So now, rather than fix any of that, the answer is to bolt AI onto a broken operating model and call it reform. You can't automate your way out of a structural under-investment in people. You get the same dysfunction, faster, with a licence fee attached and a new layer of "AI adoption" coordinators to manage it. More posts, more sign-off, more meetings about meetings, more needless work, exactly as you say. If they were serious they'd reform the pay framework and let competent people do competent work. That's harder, less photogenic, and doesn't come with an American vendor logo, which is presumably why it's not on the table.
