Bill Gates published an essay on 26 August arguing that the world is entering the AI transition without a plan, and that the decisions which will determine whether it goes well are still available to be made. For the American policy audience he is writing for, that is a fair description. For a British reader it is not, because the UK government has already answered all three of the questions he raises, and the answers are recorded with figures attached in the AI Opportunities Action Plan: One Year On, published by the Department for Science, Innovation and Technology. Five AI Growth Zones designated. £240 million for testing models. £12.5 million across the two named funds for helping regulators cope with them. And a permanent, uncapped tax deduction on precisely the capital purchases Gates wants taxed.
What Gates is actually asking for
The essay, The turbulent AI era is here. The choices we make now are critical., is long and mostly a survey: three risks (jobs, misuse, developmental harm to children), a substantial case for the benefits, and a warning that “There is no plan to ease the entry into the AI era.”
The operative part is the last third, where he sets out three things governments should do. They are worth stating precisely, because each one has a direct British counterpart that already exists.
First, build new institutions. Gates argues that “None of our current institutions were designed to handle a technology that spreads so fast and touches so many parts of our lives”, and that countries need bodies that can rank competing priorities across whole governments, paired with an international organisation modelled on nuclear inspections, aviation regulation and the ozone agreements.
Second, deliberately withhold some categories of work from automation. He calls this domain Human Reserved, borrowing the logic of a nature reserve: land where we could build, and choose not to.
Third, change the relative tax treatment of labour and capital. His formulation is blunt: “if you buy a robot, you can usually write it off right away as a business expense”, and so “The tax system nudges you toward replacing people with machines.”
Strategic Reality: Gates presents these as open questions for governments to consider. The UK has already answered two of them in the opposite direction to his recommendation, and answered the third by building an institution that measures AI rather than one that governs it.
The critical numbers
| What Gates asks for | What the UK has done | Figure |
|---|---|---|
| A cross-government body that sets priorities | AI Security Institute, evaluation-focused | £240m at Spending Review 2025 |
| Regulators equipped for AI | AI Capability Fund plus Regulators’ Pioneer Fund | £3.6m plus £8.9m |
| Binding rules on frontier developers | Promised in the 2024 King’s Speech, not introduced | No bill |
| Jobs withheld from automation | No equivalent designation | None |
| Land withheld from AI infrastructure | Opposite: five zones designated | £28.2bn investment, 15,000+ jobs |
| Tax that slows capital substitution | Opposite: full expensing made permanent | 100% first-year allowance, uncapped |
Why Britain built a measuring instrument instead of a rule
The AI Security Institute is a real institution answering a real version of Gates’s first prescription. Spending Review 2025 gave it £240 million, earmarked for three things: testing frontier models, foundational safety work, and societal resilience. A specialist recruitment function inside government has taken its technical headcount past 100, drawing people out of the major labs. By the standards of most national governments this is serious capability, and we have covered its published work repeatedly.
But an evaluation body is not the thing Gates asked for. He wants an authority able to rank priorities government-wide and, in the end, to bind behaviour. AISI tests models and publishes findings. It does not license, prohibit or compel. The body that would do those things was supposed to arrive by statute. Labour promised in its 2024 manifesto, and again in that year’s King’s Speech, to place binding obligations on the small number of firms building the most capable models. The House of Commons Library’s assessment of what happened next is one word short of blunt: legislation “has not yet been forthcoming.” The settled position instead, stated in February 2025, is that “most AI systems should be regulated at the point of use” and that “existing expert regulators are best placed to do this”.
That is a coherent policy. It is also a choice with a price, and the price shows up in the funding split. Two funds are named with figures attached for building regulators’ AI capability: an AI Capability Fund at £3.6 million, plus a Regulators’ Pioneer Fund at £8.9 million. Twelve and a half million pounds, set against £240 million for measurement. The ratio is roughly nineteen to one in favour of measuring over regulating. The plan’s wording is “including”, so the regulator side is probably larger than those two funds. It is not larger by a factor of nineteen.
Critical Context: A nineteen-to-one split toward evaluation is what a country builds when it believes the binding decisions will be made elsewhere. Britain gets excellent visibility into frontier model capability and comparatively little capacity to act on what it sees. We set out the same imbalance from the risk side in the risks 272 experts rank first, and the ones Britain is actually funding.
The exception proves the pattern. Gates notes approvingly that “Countries including Australia, the United Kingdom, and Norway are adopting protections for children online.” Britain does legislate, decisively, where the harm is visible, consumer-facing and politically legible. It declines to legislate where the harm is economic and diffuse. That is not indecision. It is a consistent rule about which kinds of damage justify statute.
The land Britain designated, and the land it did not
Gates’s Human Reserved idea is the most interesting thing in the essay and the least likely to happen here, because the UK has already run the designation exercise and designated the opposite category of thing.
Five AI Growth Zones now exist across Great Britain: a pair in Wales, one Scottish, the rest in England. The government’s own accounting puts them at £28.2 billion of investment and upwards of 15,000 jobs, helped along by planning reform and easier energy access, with discounted power for the Scottish and North East sites. An AI Growth Zone Delivery Unit exists to negotiate the electricity, the consents and the offtake deals. A handful more zones are due to be designated this year. We covered the first of these when North Wales was confirmed, and the contested arithmetic underneath the whole programme in Britain’s data centre fight runs on numbers that don’t hold.
Set beside that, the local adoption funding is £5 million per zone. Twenty-five million pounds in total, against £28.2 billion of investment the zones are meant to unlock. That works out at roughly 89 pence of adoption support for every £1,000 of infrastructure investment. The zones are designed to bring capital in, not to manage what the capital does to the local labour market once it arrives.
Procurement points the same way. From January 2026 the government has operated an AI Commercial Strategy which, in its own words, “prioritises buying from the market and innovating through challenge-led procurement”. AI Accelerator Tenders are designed to move from tender to regional-scale product in six months. An AI scan function inside the Incubator for AI provides suppliers a single front door. And the deployment is already at scale in the place where the Human Reserved argument would bite hardest: a third of chest X-rays taken by the NHS now pass through AI assistance, which the department counts as 2.4 million scans.
Reality Check: Gates asks whether we should set aside categories of work for humans. Britain has answered the question inside the NHS at a volume of 2.4 million scans a year without ever framing it as that question. The decision was made as a procurement decision, which is how most of these decisions get made.
The tax break Gates wants removed is British growth policy
This is the sharpest divergence, and the one most UK executives have not connected to the AI debate at all.
Gates’s complaint is that buying a robot can usually be written off immediately whilst hiring a person attracts payroll tax. In Britain that is not an accident of an old tax code. It is deliberate, permanent, uncapped policy. Full expensing gives companies a 100% first-year allowance on main rate plant and machinery, and HM Treasury made the relief permanent expressly to stimulate business investment. Qualifying spend is uncapped, so the larger the purchase, the larger the deduction. The Office for Budget Responsibility put a number on the switch from temporary to permanent in its Autumn 2023 forecast: an extra £14 billion of business investment across the forecast period.
So the thing Gates proposes correcting is, in the British case, a flagship growth measure with an OBR-scored investment benefit attached to it. A robot tax here would not be closing a loophole. It would be partially reversing a policy the Treasury spent political capital making permanent, at a moment when data centre and automation capital expenditure is exactly the investment the growth strategy is counting on.
Hidden Cost: Every AI Growth Zone business case, every automation capex model and every data centre investment appraisal in Britain currently assumes full expensing. That assumption is load-bearing across the whole programme, which is precisely what makes it hard to change and worth understanding before you build a five-year plan on top of it.
What this means depending on where you sit
| Who you are | What the foreclosure means | What to do about it |
|---|---|---|
| Large UK employer | No Human Reserved protection is coming; workforce transition is entirely your policy | Write your own displacement rules now, before a restructure forces them |
| SME | Full expensing is a genuine, permanent advantage on automation capex | Use it deliberately rather than accidentally; model the tax treatment explicitly |
| Public sector buyer | Procurement is the actual site of AI policy, not legislation | Treat contract terms as the governance layer, because nothing above them binds |
| Regulated firm | Your regulator sets your AI rules, and is funded from a small pot | Read your regulator’s published AI plan; do not wait for a cross-sector Act |
| Investor or developer | Growth Zone commitments are the least reversible part of the picture | Discount policy-reversal risk on infrastructure, price it on tax treatment |
The human factor here is worth naming directly. Gates writes movingly about the caregivers who looked after his father, and about the 55-year-old construction worker you cannot simply redirect into elder care. That argument has purchase in Britain too, but there is no institution here whose job it is to act on it. The Growth Zone Delivery Unit brokers power and planning. AISI tests models. Sectoral regulators handle point-of-use harm within their remits. Nobody owns the question of what happens to the people, which is the same structural gap we identified in UK government AI projects keep stalling: the failure is rarely the technology, and usually the absence of an owner.
What to do while the choices stay closed
The useful move is not to campaign for Gates’s framework. It is to work out which British commitments are genuinely reversible, on what timescale, and to plan accordingly.
Reversible within one Budget: the tax treatment. Full expensing can be narrowed or carved out at any fiscal event. It would be costly and contested, but it is a stroke of the pen. If your automation business case only works with 100% first-year relief, it carries real political risk.
Reversible per contract: procurement terms. Nothing in current policy stops a UK buyer writing human-review requirements, redeployment commitments or exit rights into a contract. Where there is no statute, the contract is the statute. This is the single largest piece of unused leverage most organisations hold.
Effectively irreversible: the Growth Zones. Designation, planning reform, energy access and private capital commitments at £28.2 billion do not unwind on a policy change. Plan on the assumption that this infrastructure gets built.
Implementation Note: Sequence your response to match that reversibility ladder. Contract terms this quarter, because they are yours to write. Tax exposure this year, because it is a live political variable. Infrastructure assumptions permanently, because they are settled.
If you are early in AI adoption: map which of your roles are exposed to procurement decisions made above you, particularly in regulated or public-sector supply chains. The decision that displaces work in your organisation may be taken by your client, not by you.
If you are mid-maturity: audit your automation capex against the full expensing assumption, and write your own displacement policy. No external rule will supply one.
If you are advanced: engage your sectoral regulator directly on its published AI plan. Regulators are working with modest AI budgets and are visibly short of practitioner input. A firm that shows up with evidence has disproportionate influence on rules that do not yet exist.
Four things this analysis does not solve
The evaluation-to-action gap has no owner. AISI can determine that a model is dangerous and has no mechanism to stop its deployment. Mitigation: assume nothing you procure has been cleared for use in your context, and build your own acceptance testing.
Point-of-use regulation fragments accountability. When every sectoral regulator writes its own AI plan, a firm operating across sectors faces overlapping and possibly inconsistent expectations with no arbiter. Mitigation: maintain one internal AI control framework and map it to each regulator, rather than the reverse.
Growth Zone benefits and Growth Zone costs land on different people. Investment counts nationally; grid pressure, water use and planning displacement land locally. Mitigation: if you are siting in or near a zone, treat local consent as a project risk with a duration, not a formality.
A tax change would arrive with no notice. Fiscal events do not consult on capital allowances in advance. Mitigation: stress-test any multi-year automation plan against writing-down allowances rather than full expensing, and know what breaks.
The takeaway
Gates is right that the AI transition needs a plan and mostly does not have one. Where his essay is misleading for a British audience is in the tense. He writes about choices to be made. Britain made them: an evaluation institute rather than a regulator, designated land for compute rather than protected categories of work, and a permanent tax preference for machines over payroll. None of these was announced as an AI labour policy. Together, that is exactly what they are.
Three things follow. Contracts are the governance layer, because nothing above them binds. The tax treatment is the one genuinely open variable, so know your exposure to it. And no institution currently owns the human side of the transition, which means it is yours to own.
Strategic Insight: The question is not whether you agree with Gates. It is whether your five-year plan quietly assumes a British policy environment that has already been decided against it.
Your next steps
Immediate actions (this week):
- Identify every AI or automation commitment in your plan that assumes full expensing
- Locate your sectoral regulator’s published AI plan and read it
- List the AI-related decisions being made about your organisation by your clients or suppliers
Strategic priorities (this quarter):
- Write human-review and redeployment terms into your next AI contract
- Draft an internal displacement policy, since no external rule will supply one
- Model your automation capex under writing-down allowances as well as full expensing
Long-term considerations (this year):
- Build your own acceptance testing rather than relying on any external clearance
- Map one internal AI control framework across every regulator that touches you
- If siting near an AI Growth Zone, price local consent as a timed project risk
Source: The turbulent AI era is here. The choices we make now are critical. by Bill Gates (Gates Notes, 26 August 2026). UK figures are drawn from DSIT’s AI Opportunities Action Plan: One Year On, the House of Commons Library briefing on AI regulation in the UK, and HM Treasury’s policy paper on permanent full expensing. Gates’s citation of Pope Leo XIV’s encyclical on AI is one we examined at the time in The Pope, Anthropic and Westminster.
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether the promised AI legislation appears in the next King’s Speech, because that is the one event that would reopen a choice Britain has effectively closed. Read more analysis at Insights, or get in touch.