Double Vision. AI as Co-Pilot or Autopilot
Why clients need more than AI literacy, they need improved humans to get the most from AI
There’s a particular kind of politician who can be in the same room as an uncomfortable truth for years without ever quite acknowledging it. Rishi Sunak, on the subject of artificial intelligence and the job market, appears to be the poster-child for willfully ignoring the obvious until it suits him.
In the autumn of 2023, as Prime Minister, he convened the world’s great and good at Bletchley Park and told us, variously, that AI would be a “co-pilot” rather than a replacement, that workers should not be worried, and that the government’s job was to provide a “world-class education system” so that everyone could ride the wave. Last month, as a private citizen and paid adviser to Anthropic and Microsoft, he went on BBC Newsnight and told us that entry-level hiring is flattening, graduates’ concerns are justified, and we should probably abolish National Insurance.
The same man. Two and a bit years apart. Broadly the same technology. Two quite different stories.
We want to take this seriously … not as a gotcha, though there is a gotcha available if you want one, but as a case study in how the conversation about AI keeps going wrong. Because Sunak isn’t an outlier here. He’s doing what nearly everyone in this debate does, which is to look at a very large, very cross-cutting phenomenon through exactly one lens at a time and then be surprised when the picture keeps changing shape.
Let’s be fair to the 2023 version first. At Bletchley, and in his Royal Society speech the week before, he landed on a framing that was genuinely popular in tech-adjacent policy circles: AI as augmentation. The welfare caseworker gets their paperwork done faster. The lawyer drafts contracts in minutes. The doctor spots the retinal disease earlier. Co-pilot, not autopilot. Nobody loses their job; everyone gets better at the one they have. (This of course assumes organizations are not profit maximizing entities and wouldn’t see an opportunity to cut labor costs)
When pressed on labor-market disruption, he gave the answer that British politicians have given about every technological transition since the spinning jenny, which is: education. We will train our way out of this. A world-class education system will deliver a workforce resilient to whatever comes. Creative destruction will be, on net, creative.
The trouble (and this was pointed out at the time, loudly, by people like Carissa Véliz at Oxford) was that the framing of the invited coterie sounded suspiciously close to what the companies building the technology wanted governments to say. Don’t regulate us; train your workers instead. It was also, notably, the framing that allowed a Prime Minister to host a global AI summit without having to promise anything awkward about jobs.
And sitting next to him on stage, Elon Musk was telling anyone who’d listen that eventually no job would be necessary at all. So it’s not as though the more disruptive story was hidden. Sunak just chose not to dwell on it.
Fast-forward to last month. The co-pilot has developed a worrying habit of flying the plane. Sunak told Newsnight that CEOs are privately telling him “flat is the new up” — meaning they think they can grow their businesses without hiring more people, because AI is absorbing the work. Entry-level roles in law, accountancy and the creative industries are being compressed. Graduates who were told in 2023 not to worry are, in 2026, being told their worry is justified. And even now, we suspect that Sunak has again pulled on his rose-tinted glasses.
His proposed fix has also moved. Education is still there — rebranded as “AI literacy,” which he charmingly calls “the driving licence for the modern workforce.” But there’s a new structural piece: abolish National Insurance over time and replace it with taxes on corporate profits, on the grounds that as AI shifts value from labor to capital, the tax base should follow.
This is a more interesting policy than the headlines gave it credit for. It is also, as it happens, a policy that is unusually gentle on the firms driving the disruption — you tax the upside, you don’t restrain the technology, and the shift happens “over time.” If you were, hypothetically, an adviser to two of those firms, it is not the worst policy you could be advocating.
But of course, taxing corporate profits is anything but straightforward. Just pinning profits to a geography so that they can be taxed is akin to nailing jelly to a wall … and that is before other governments start threatening retaliation.
https://www.euronews.com/next/2026/04/24/explained-what-is-the-uk-digital-services-tax-and-why-has-it-angered-trump
So, has Sunak changed his mind, or has his mind changed positions?
The charitable reading is that the evidence is becoming clearer. The FT-Focaldata numbers, the Accenture survey showing half of UK executives now expect AI to reduce headcount within a decade, the AWS adoption data — none of that was on the table at Bletchley. A reasonable person updates when the data arrives.
The less charitable reading is that his diagnosis has tracked his incentives with unusual fidelity. As PM courting investment, he needed to reassure voters and flatter the sector. As an adviser paid by that sector, he needs the sector to be seen as a net good, but one whose externalities merit tax accommodation rather than regulatory pushback.
Both of those readings can be partly true. The awkward bit is that Sunak hasn’t acknowledged the shift. There is no “I got this wrong at Bletchley and here’s why.” There is only the new diagnosis, delivered with the same confident tone as the old one, and no bridge between them.
And here is where we want to widen the lens because Sunak is not the story, but rather a symptom. The AI labor-market debate has been dominated, at every stage, by single-lens analysis. The technologists talk about capabilities. The economists talk about aggregate productivity. The politicians talk about education. The ethicists talk about bias. The tax people talk about tax. The Musks of this world talk about post-work utopias. And each group is right about its bit (except maybe the bit about a post work utopia), and collectively they are producing a picture that keeps missing what the thing actually is.
A polymathic reading of the Sunak arc goes something like this. You can’t diagnose AI’s effect on junior roles without simultaneously understanding: (1) the micro-economics of why firms hire entry-level in the first place, which is partly about building a talent pipeline and partly about doing unglamorous work cheaply; (2) the technology’s current capability curve, which happens to hit exactly that unglamorous-work tier; (3) the labor-market institutions (apprenticeships, professional qualifications, graduate schemes etc.) that were built around that tier and are now being hollowed out; (4) the tax structure, which subsidizes capital investment and taxes employment; (5) the education system, which is still training people for the jobs that are disappearing fastest; and (6) the political economy, in which the people best placed to see the shift are the same people least incentivized to talk about it plainly.
No single discipline holds all six. And if you’re missing any of them, you end up either with 2023 Sunak (”don’t worry, education will handle it”) or 2026 Sunak (”actually, worry, and let’s restructure the tax code”), which are both reasonable moves within their lane but neither of which actually answers the question.
A PolymathMind call to arms
Where does PolymathMind stand? We’re not here to tell you National Insurance should stay or go, or which party has the better AI white paper. That’s not our competence. What we do think is this.
1. AI is going to be hugely disruptive to work and careers. Not in the hand-wavy “changes in the labor market” sense that politicians reach for when they want to acknowledge something without committing to it. In the specific, structural sense that the entry-level tier of knowledge work is being compressed right now, that the apprenticeship logic underpinning whole professions is breaking, and that the career ladder a twenty-two-year-old graduate is being handed in 2026 looks nothing like the one their manager climbed.
2. Education needs an overhaul. Not tweaking. Not another strategy document. The system we have was designed to produce the workers that firms used to hire, doing the work that AI now does faster and cheaper. Telling school leavers to study harder for a set of jobs that are being hollowed out is not a plan, it’s dereliction! And “AI literacy” as a driving-licence metaphor doesn’t cut it — knowing how to prompt a model is the floor, not the ceiling.
3. And platitudes aren’t going to get us there. “Embrace the future.” “Lifelong learning.” “Humans and AI working side by side.” Cliches! They let people feel like they’ve said something without committing to anything.
AI-mediated work is a form of collective intelligence, and the collective only performs if the human side is upgraded as seriously as the machine side.
The same firms pouring billions into models, GPUs, and data pipelines are spending a rounding error, by comparison, on the human capability that determines whether any of it produces value. Judgement under uncertainty. The ability to reframe a problem when the model confidently answers the wrong question. The editorial instinct to know when the first draft is good enough and when it isn’t. Domain knowledge deep enough to spot when the AI is hallucinating plausibly. These are the capacities that make AI-augmented work actually pay off, and they are exactly the capacities that the current training pipeline — from schools to graduate programs — is not deliberately building.
We should be honest about why this matters, and to whom. Businesses are not charities. They won’t invest in upgrading their people out of philanthropy. They’ll invest when it becomes obvious that the firms that pair strong AI deployment with strong human capability are eating the lunch of firms that just bought the licences.
The disruption is real and already underway, the response so far has been inadequate, and the most interesting work ahead is in raising the human game — not defending it, not lamenting it, raising it — so that the collective intelligence of humans-plus-AI is worth more than the sum of its parts.
That’s the PolymathMind focus.
If you like the PolymathMind, recommend it to a friend
Executive Summary
“Flat is the new up” — CEOs growing revenue without growing headcount — is now the operating reality.
“AI literacy as the modern driving licence” and tax restructuring miss the systemic question: what happens to apprenticeship models, graduate schemes, and the institutions built around entry-level work.
Technologists, economists, politicians and ethicists each see one piece. Nobody is holding labour-market structure, education, tax, and capability trajectories in the same frame — which is why the framing keeps changing shape.
Value lands where strong AI deployment is paired with strong human capability — judgement, taste, domain depth, editorial instinct. The human side needs investment as serious as the spend on models and infrastructure.
Businesses will upgrade their people because the firms that do will outcompete those that just buy the licences. The call to arms is to raise the human game — clear-eyed, no platitudes.