AI 2036

This is sizrok’s own view of where AI actually takes UK business over the next ten years. Not a summary of what the industry’s loudest voices are currently saying about themselves.

Three things anchor it. AI reshapes far more jobs than it deletes, and not having basic AI skills becomes as costly as not knowing how to use a computer did twenty years ago.

The UK ends up back inside the EU well before 2036, because the economics of staying out get harder to justify every year that passes. And none of this plays out as the flattened, jobless, ownership concentrated future some of the industry’s biggest names keep describing, because the physical and political constraints this piece lays out don’t allow it to move that fast.

Here’s the case for all three, in full.

A note on timing. Parts of this piece, particularly on UK politics, were overtaken by events while we were writing it. We’ve left the original argument in and updated it in place, because the update turned out to prove the point rather than undercut it.

What is AI 2036

AI 2036 is sizrok’s ten year forecast for how artificial intelligence actually reshapes UK business, built on where compute, regulation and investment stand today, not on speculation or on the say so of whoever benefits most from a given prediction.

It covers what happens to jobs, why AI literacy is about to matter the way computer literacy did, how the UK’s regulatory and European position develops, why London and Manchester are becoming the country’s two AI cities, and why the loudest voices in the industry don’t speak for where it’s actually heading.

Every figure in this piece is sourced and dated, not guessed.

The jobs question: fewer jobs is the wrong frame

Let’s deal with the uncomfortable part first.

Some jobs genuinely go. Repetitive processing, first line triage, high volume drafting, the kind of work that was always a stopgap between two decisions rather than a decision itself.

Pretending otherwise doesn’t help anyone plan for 2036.

But “fewer jobs” is the wrong frame for what’s actually happening, because it treats every role as a single, disappearing unit. It isn’t.

Break the jobs question down properly and three tiers appear.

Three tiers, not one blob

Tier one is roles that get fully automated. These are narrow.

Single step, low judgement, high volume tasks, the kind of work a spreadsheet macro would have taken twenty years ago if the input data had been tidy enough. This tier is real, and it’s smaller than the headlines suggest.

Tier two is roles that get reshaped. A person still does the job.

They just spend the hours a machine used to eat on the parts a machine still can’t do, the judgement calls, the client relationships, the exceptions that don’t fit the pattern. Most jobs sit here.

Tier three is roles that get created. Someone has to map the workflow before anything gets automated.

Someone has to maintain the integration when the underlying software changes its API without warning. Someone has to review the edge cases the model gets wrong, and someone has to explain the system to a regulator or a client who wants to know what happens when it fails.

None of this existed as a job category five years ago. All of it scales with adoption, not against it.

Most of the noise in the public AI debate comes from conflating tier one with the other two, and it’s worth naming that plainly.

The historical pattern

Every general purpose platform shift, the web, mobile, cloud, displaced roles and created a larger number of new ones downstream, usually with a multi year lag between the two. AI compresses that lag because the tooling to build on it is cheaper and faster to learn than any prior platform.

Someone can go from knowing nothing about AI agents to shipping a working one inside a fortnight, something that took months on the last three platform shifts.

Take a UK example. A 40 person recruitment agency doesn’t shed its consultants when it automates CV screening and first contact.

It reallocates their time toward the parts of the job that were always the actual value, the calls, the persuasion, the judgement about fit. Fewer processors, more integrators, and a whole tier of small agencies that didn’t exist three years ago, reducing repetitive admin so the people who were doing it can do something a machine still can’t.

The mechanism matters more than the headline. Workflow automation doesn’t remove the need for people who understand the workflow.

It removes the need for people to manually execute it, which is a different thing entirely, and the gap between those two facts is where most of the public argument goes wrong.

Like learning to use a computer, not like losing your job

There’s a comparison worth making plainly, because it’s the one that actually helps someone plan rather than panic.

AI is not going to replace you. Not knowing how to use it is going to become the same kind of gap that not knowing how to use a computer became somewhere around the turn of the century.

Nobody lost their job in 2001 for being unsure around a computer. But within a decade, not being comfortable with email, spreadsheets and the basic software your role ran on stopped being a quirk. It became a real limit on what jobs you could hold, and how far you could go in the ones you already had.

AI is running the same curve, faster. It won’t replace a good account manager, a good site foreman, or a good bookkeeper.

It will quietly become the baseline expectation that they know how to use it, the way a laptop and an email address already are. That’s what tier two, reshaped roles, actually means in practice. The judgement stays human, the literacy requirement around it is new, and it’s arriving faster than the last one did.

The compute ceiling

This is the strongest argument in this piece, and it’s the one built entirely on evidence rather than opinion.

Start with the headline number. Roughly 30 to 50% of planned 2026 US data centre capacity is expected to be delayed or cancelled.

Of the roughly 12 gigawatts planned for 2026, only around a third is currently under active construction, according to reporting corroborated across Bloomberg, TechRadar and Tom’s Hardware.

The bottleneck has moved. It used to be GPUs.

Now it’s transformers, switchgear and grid connection approval, the unglamorous physical infrastructure that takes 24 to 36 months to clear in major markets, regardless of how much capital is sitting behind the project. Gartner projects that 40% of AI data centres will be power constrained by 2027.

None of this is a funding problem. Hyperscalers are still committing more than 650 billion dollars in 2026 capital expenditure on AI infrastructure.

The money exists. The physical capacity to deploy it doesn’t, not on the timeline the “AI replaces everyone” narrative needs.

Why this caps jobs, not just growth

Here’s the detail that makes this argument bite rather than just sitting there as a supply chain trivia aside.

Full workforce automation isn’t a training problem. It’s an inference problem.

Training a model happens once, in a data centre, over weeks. Running it for millions of daily tasks across an entire economy needs sustained, live compute, every day, for every company that wants it, at the same time.

That’s the part that needs the power stations, the transformers and the grid connections currently stuck in an approval queue.

There’s a second, quieter brake sitting underneath the physical one. Scarce compute doesn’t just slow deployment, it prices it.

When inference capacity is constrained, providers ration it by price. That keeps full scale automation expensive for longer than the “AI is basically free now” narrative assumes.

Put those two together and the ceiling gets clearer. Full scale automation of a workforce needs sustained inference capacity, at scale, for every company that wants it, simultaneously.

That capacity doesn’t exist yet, and won’t for years on current build rates. This is a genuine, physical brake on the pace of displacement, independent of anyone’s ethics or intentions.

Worth being honest about what this brake actually does over a ten year window, rather than overstating it. It caps the pace, not the destination.

The 2026 to roughly 2030 stretch this piece keeps returning to genuinely can’t look like the accelerationist story, the infrastructure isn’t there and won’t be built in time. By the early 2030s, once the current multi hundred billion dollar wave of data centre construction actually finishes clearing its grid connection queues, a lot of that physical constraint eases.

Which means the second half of this forecast, 2031 to 2036, isn’t decided by physics any more. It’s decided by who owns what gets built and who’s allowed to regulate it, which is exactly what the rest of this piece is actually about.

Worth flagging one more technology thread running in parallel, without overstating it either. Quantum computing is maturing on a similar decade long horizon, and it doesn’t solve today’s GPU and grid bottleneck, current quantum hardware isn’t a substitute for the classical compute AI training and inference actually run on.

But it’s a second frontier technology businesses will want at least a working literacy in before 2036, on top of AI, not instead of it.

Which is exactly why the businesses moving first are focused on workflow agents built for one specific bottleneck, not a platform promising to replace a department overnight. The infrastructure to support the second kind of promise doesn’t exist yet.

The infrastructure for the first kind already does.

Regulation: the fork in the road

If compute is the physical brake, regulation is the political one, and this section changed while we were still writing it.

Where things stood a month ago

For most of 2026 the answer was simple, and slightly depressing. The UK had no standalone AI law.

Labour promised binding legislation on the most powerful models in the 2024 King’s Speech and reaffirmed the commitment since. It was never introduced.

AI in the UK was governed through existing regimes instead, UK GDPR, the Data (Use and Access) Act 2025, and sector regulators like the FCA, the MHRA and Ofcom applying their own rules within their own remits.

In May 2026 the government’s King’s Speech briefing notes announced something different: a Regulating for Growth Bill, built around regulatory sandboxes under an “AI Growth Lab” that let specific rules be temporarily relaxed for licensed pilots in sectors including healthcare, professional services, transport and manufacturing.

Critics in the House of Lords called this a pro innovation pivot away from the binding approach originally promised, not an extension of it. That reading was hard to argue with at the time.

The EU, for context, hadn’t waited for any of it. Its AI Act is binding law already, arriving in stages.

Prohibited practice rules are active now. Transparency obligations apply from 2 August 2026. Watermarking requirements for AI generated content follow on 2 December 2026.

Then the fork actually turned

Keir Starmer resigned as prime minister on 22 June 2026, after eighteen months of policy reversals, a run of local election defeats and falling poll numbers. Andy Burnham, mayor of Greater Manchester since 2017, was fast tracked into a vacant safe Labour seat, won the leadership unchallenged, and was sworn in as prime minister on 20 July 2026.

In a country where the ruling party can change its leader, and therefore its prime minister, without a general election, that’s exactly what happened.

This isn’t a hypothetical mechanism any more. It’s last week’s news, and it lands directly on the argument this section was already making before it happened.

A government’s regulatory posture is a function of who’s in charge, and who’s in charge changed inside a single news cycle.

Burnham brings a genuinely different starting position on AI to Downing Street. As mayor he refused to award Palantir Technologies a single contract across his near decade in office, a decision Whitehall watchers now read as an early marker of how his government will treat the more aggressive end of the US AI industry.

He has publicly called for tighter regulation of AI and Big Tech, arguing that four decades of deregulation did not, in his words, trickle down very much at all. Advisers briefed the Financial Times that his incoming AI strategy centres on three things: British ownership of AI infrastructure including data centres, tech sovereignty, and protecting UK jobs from AI displacement, explicitly moving away from what his own team calls an overly US centric approach.

None of that is enacted policy yet. It’s a direction, days old, and it has already drawn a backlash from parts of the UK tech sector who would rather keep the sandbox model described above.

But the direction itself matters for a ten year forecast more than the detail does this early. A prime minister who spent a decade running the AI ready city this piece already argued would matter, and who used that decade to say no to Palantir, is a materially different starting point than the government that wrote the Growth Bill in May.

Our read on it

We think this direction holds, more often than not. A light touch, growth first approach is a function of whoever’s priorities are currently in charge, and it just stopped being a settled thing mid cycle, without anyone needing to wait for the next election.

A government reviving the shelved 2024 legislation, adding worker protections, mandatory human in the loop rules for high stakes decisions and proper audit trail requirements, was always a plausible outcome rather than a fantasy. It’s a more plausible one this week than it was last month.

It could still stall. Governing is harder than campaigning, a Treasury facing 0.8% growth doesn’t hand out easy wins, and a backlash from parts of the tech sector is already forming.

We’re naming that honestly rather than pretending the direction is guaranteed. But whichever way it lands, the EU’s calendar from the section above doesn’t move, and businesses serving EU clients need to plan for it either way.

Britain’s position

Two questions sit under this heading, and we’ll give you sizrok’s actual view on both, not just the safe, hedged version.

The government that set the UK’s current red lines on Europe, no return to the single market, no customs union, no freedom of movement, was Keir Starmer’s. Starmer resigned as prime minister on 22 June 2026.

Andy Burnham took office on 20 July 2026. Those red lines were Starmer’s own positioning, not a law, and a new prime minister with a different political instinct is under no obligation to leave them exactly where his predecessor did.

Here’s our actual prediction, not the hedged one. We believe the UK rejoins the EU, in substance and most likely in name, well before 2036.

Not because any government has said so yet. Because the economics keep getting harder to argue against.

Every year outside the single market is another year of friction cost on services trade, the sector that actually carries the UK economy. 0.8% GDP growth in 2026 isn’t a one off, it’s the shape of what staying out looks like when you let it run for a decade.

Rejoining isn’t a favour to Brussels. It’s what services trade needs to grow again, what makes cross border ecommerce genuinely frictionless rather than paperwork heavy, and what brings back the businesses that quietly built an EU footprint elsewhere after 2016 rather than deal with the customs and data adequacy overhead.

That’s our actual forecast for 2036. Services trade opens back up properly. Ecommerce accelerates because the friction that’s been taxing it since 2021 disappears.

Businesses that left, or that never set up here in the first place because of the overhead, come back or arrive for the first time. It compounds with the AI led agency boom this piece covers later, rather than sitting apart from it.

The alternative is the one nobody in government wants to say out loud yet. If it doesn’t happen, the drift continues, growth stays where it is, and the gap between the UK and its nearest, richest trading partner keeps widening rather than closing.

It’s a matter of time either way. Our bet is on the first version.

What was already booked before the leadership changed points the same direction, even if nobody’s calling it that yet. The UK rejoined Erasmus+ in 2025.

A second UK EU summit was scheduled for 22 July 2026 in Brussels, two days after Burnham’s swearing in, covering trade, security, energy and youth mobility. Talks continue on the SAFE defence programme and on integrating the UK into the EU electricity market.

The regulation gap from the last section adds to the case rather than sitting apart from it. With the EU AI Act’s binding transparency and watermarking rules landing in August and December 2026, UK businesses serving EU clients already have to comply with EU rules regardless of what Westminster decides at home.

That’s the practical shape of alignment arriving years before the political label catches up to it. An AI automation agency working across the UK should build for the EU’s risk based approach either way, because on our forecast, it stops being optional.

London and Manchester are becoming the UK’s two AI cities

This is the section where the national picture gets local, and where the numbers are genuinely striking.

The numbers

Greater Manchester’s AI sector is valued at 4.7 billion dollars in 2026, up 9% year on year, according to Invest Manchester’s own data. The region has attracted almost a billion dollars in cumulative AI investment since 2010 and is now home to 13,500 AI professionals.

Manchester has topped the SAS AI Cities Index as the UK’s most AI ready city outside London for three years running.

That’s not a fluke. It’s structural.

London holds the capital markets, the enterprise headquarters and the American tech companies opening their UK offices. Manchester holds a genuine research base along the university corridor, a lower cost of operation, and a growing density of AI specific talent that a smaller agency can actually hire against without competing head to head with London salaries.

That combination, research plus cost plus a deliberate regional investment push, is what makes a city genuinely AI ready rather than just AI adjacent.

By 2036 expect a genuine two city model. London for capital and headquarters, Manchester and the wider North West corridor for delivery, talent density and the lower cost base that lets agencies actually build the automations rather than just sell them.

Then Manchester’s mayor became prime minister

This section was largely written before the news below happened. We’re leaving the original numbers and argument exactly as they were, and adding what happened next, because it’s better evidence for the case than anything we could have written ourselves.

Andy Burnham was mayor of Greater Manchester from 2017 until his election as prime minister in July 2026. For nearly a decade, the city this piece has spent several paragraphs arguing matters had a mayor who built its AI and Data Innovation Office, backed Health Innovation Manchester’s public sector AI work, and consistently refused to award Palantir Technologies a single contract, a company Sadiq Khan’s Metropolitan Police also blocked a deal with in May 2026.

That’s not a small detail for a piece arguing Manchester matters to UK AI policy. It’s the strongest possible version of the argument available.

The country’s approach to AI regulation, tech sovereignty and the more aggressive end of the US AI industry is now set, at least in its first weeks, by the person who spent a decade building the model city this piece already picked out, months before he took the job that lets him apply it nationally.

Whether “Manchesterism”, the label commentators have already attached to his platform of devolution and reindustrialisation, actually survives contact with a Treasury, a fractious cabinet and an economy growing at 0.8% is a genuinely open question, and this piece isn’t claiming otherwise. What isn’t open to argument is the symbolism.

By 2036, expect the London and Manchester two city model this section already predicted to be reinforced by national policy as much as by investment figures, because the person setting AI policy nationally spent the last decade deciding what that policy should look like in one specific place first.

Beyond the obvious two

The pattern beyond these two cities is still worth watching. The same combination that built Manchester’s position, a research base, a lower cost of operation and a deliberate regional push, could plausibly land in Leeds, Bristol or Edinburgh before 2036.

Nothing certain yet. Worth watching regardless, and worth watching a little more closely now that the model has a direct line to Downing Street.

The boom nobody’s pricing in

Two previous booms are worth remembering, because AI is compressing both of them into a single moment.

The dot com boom taught every business it needed a website, then that it needed an app. The social media boom taught every business it needed a presence, then a strategy, then an agency to run the strategy because the platforms changed faster than anyone could keep up with in house.

AI compresses both lessons into one. Every business now needs an integration, an agent, a workflow, and increasingly, an agency to build and run all three because the tooling changes every quarter, not every few years.

Why this needs agencies, not just software

Here’s the part that’s easy to state as an assertion and harder to actually justify, so it’s worth doing properly.

Off the shelf AI tools solve generic problems well. They solve a specific business’s actual bottleneck badly, because that bottleneck is shaped by that business’s own systems, exceptions and history, not by whatever a SaaS product assumed when it was built for everyone at once.

That gap, between generic tooling and one specific operation, is exactly the gap an agency fills. It’s structurally the same gap web agencies filled in 2001 and social agencies filled in 2011.

Every business is about to have an AI guy

Here’s where the historical pattern gets specific, and it’s worth walking through properly, because the phrase “AI guy” gets said as a joke and it shouldn’t be.

Every general purpose technology shift produces the same social pattern. A business doesn’t understand the new thing well enough to run it without a person whose whole job is understanding it, so it hires or designates one.

The 1990s produced the webmaster, one person who understood HTML and FTP well enough to be the only one allowed near the company website. The 2000s produced the IT manager, then the systems administrator, as networks and email got too complicated to leave to whoever was good with computers.

The 2010s produced the social media manager, then the head of digital, as platforms multiplied faster than any generalist could track. Each of these roles started as a joke title at a handful of companies and ended as a standard line on an org chart within about five years.

AI is running that same pattern at a faster clock speed, and there’s already hard data on it. IBM’s Institute for Business Value found that 76% of organisations surveyed in 2026 have a Chief AI Officer, up from just 26% in 2025.

That’s not gradual adoption, that’s a role going from rare to standard inside a single year, and IBM’s own data shows companies with a CAIO see a real, measured return, roughly 5% higher return on their AI spend than companies without one.

The number that matters more for the businesses sizrok actually works with isn’t the large enterprise figure, it’s the shape underneath it. Full time Chief AI Officers cost real money, in the US market where the data is most complete that’s roughly 250,000 to over a million dollars a year, and that cost only makes sense once a company is large enough and AI central enough to justify a dedicated executive.

For a business well under that scale, the emerging standard isn’t a full time hire at all, it’s a fractional one, someone doing the same job for eight to fifteen hours a week at a fraction of the cost. That’s a genuinely different job to the one the headlines describe, and it’s the one that’s actually going to exist inside most UK SMEs by 2036.

Worth naming the trap here as plainly as the opportunity, because it’s the kind of detail that separates a business making a real decision from one making a defensive one. Some companies hire a Chief AI Officer purely as a signal, a line for the annual report and an answer for the next earnings call, with no budget and no engineers behind the title.

Eighteen months later they own a strategy deck and not one AI feature actually running in production. The market reads through that within a year. A title isn’t a capability, and the businesses treating it as one are the ones still explaining themselves at the next board meeting.

None of this points toward fewer jobs overall, worth restating plainly given the section this sits inside. Robert Half’s 2026 survey found that 41% of small business leaders expect AI adoption to produce a net increase in jobs over the next two years, not a decrease.

The roles feeding that increase already have names and salaries attached, not just Chief AI Officer but AI engineers, MLOps specialists whose job listings grew almost tenfold across five years on LinkedIn’s own tracking, and the workflow automation specialists who map a business’s actual process before anyone touches a model.

We build this for a living, so we’ll say it plainly rather than pretend otherwise: the services we actually run sit in the second half of this boom, not the first. The tools exist already.

Someone still has to wire them into a specific business, and that’s a service, not a product, however many companies try to sell it as the latter. A live example, an inbound receptionist agent we built recently, does exactly this, sitting inside one business’s actual phone workflow rather than replacing anyone who worked it before.

It’s the fractional model described above, applied to one specific bottleneck rather than an entire function.

What “AI jobs” actually means by 2036

The phrase gets used vaguely, and it shouldn’t be. Break it into what it actually means.

People who scope and build automations, agencies and in house builders alike. People who maintain and audit them once they’re live, a genuinely new operations discipline that barely exists today.

People who sell and support them, a new tier of account management built around technical fluency rather than a script.

None of these three categories existed as job titles five years ago. All three scale directly with adoption, and adoption is still a small minority of UK SMEs today, which is the actual size of the opportunity most forecasts underrate.

Marketing agencies are adopting this fastest of any sector we work with, for the obvious reason that they already sell speed and iteration for a living.

Why “nobody will work” is nonsense

Some prominent voices in the industry describe a future of abundance where AI does the work and a universal income or dividend covers everyone regardless. It’s worth stating that claim honestly before knocking it down, because it deserves a real argument, not a strawman.

The rebuttal has two legs, and both were built earlier in this piece rather than asserted here for the first time.

Leg one is the compute ceiling. The “everyone stops working” scenario requires inference capacity that doesn’t exist on any credible near term timeline.

You cannot automate an entire economy’s labour with data centres that are 30 to 50% behind schedule and power grids running 24 to 36 month connection queues. The physics comes before the philosophy.

Leg two is who owns the gains. Even where full automation eventually becomes technically possible, the gains accrue to whoever owns the models and the compute, not to the people displaced, unless regulation and ownership structures change.

Left unmanaged, that’s a real risk worth naming plainly: a small ownership class capturing the productivity gains of a technology built on public data, public infrastructure and public electricity grids.

We don’t think that’s where this actually lands, and we’ve already shown our working on why. Regulation is already moving, not in theory, but in a UK government sworn in three weeks ago on a platform of exactly this kind of pushback.

The distinction between AI and automation matters here too, because automation has always been a tool that amplifies whoever controls it, and this decade is the one where who controls it gets decided.

That’s not inevitable doom. It’s an open question with a government, a regulatory calendar and a set of ownership structures already starting to answer it, and our forecast is that they answer it reasonably well.

Democracies have course corrected on concentrated power before, more than once, when enough people decided it mattered. Nothing about this decade makes that harder than it’s ever been. If anything, the correction already started earlier than most people expected.

The loudest voices are not the whole industry

A handful of founders dominate the AI conversation, and it’s worth separating what they say from what the industry actually does, and from what sizrok actually believes happens.

Sam Altman has spent 2025 and 2026 arguing publicly for both faster deployment and for regulation, while OpenAI has lobbied against several of the specific rules being proposed. That’s a tension worth naming rather than taking at face value.

Elon Musk owns the platform much of this conversation happens on. In the past year he’s launched a political party, largely paused it, and shifted to funding the very party he broke from ahead of the 2026 midterms.

He’s also put a specific date on his own prediction for how this all plays out. He told The Economist in July 2026 that he expects AI to exceed total human intelligence by around 2031, and for humans to lose control of its direction by 2036, the same year, as it happens, this piece is named after.

He rates the odds of a genuinely catastrophic outcome at 10 to 20%, and thinks the companies building it should check their own work rather than have elected governments do it. That interview aired the same week SpaceX passed a 2 trillion dollar valuation, making him the world’s first trillionaire.

Our view, plainly: someone estimating a one in five chance of catastrophe, asking not to be regulated by anyone but himself, and profiting more than any person in history from the outcome either way, isn’t who any of us should be taking direction from. His forecast might be right about the technology. His judgement about who should be trusted to check it is the judgement of someone with a great deal riding on nobody checking too closely, which is most of what this piece has been arguing against from the start.

Peter Thiel has been open for years about preferring markets with less democratic interference. That’s a legitimate position to hold, and a fair one to disagree with. We do.

It’s also worth noting that Thiel’s Palantir, arguably the clearest commercial expression of that worldview inside the AI industry, spent nearly a decade failing to win a single contract from the mayor of the one UK city this piece has spent several sections arguing matters most, and that mayor now runs the country.

None of this is a reason to distrust the technology itself. It’s a reason to build for the businesses actually using it, not for the handful of men currently arguing about whose vision of the future gets to matter most.

Hollywood’s own verdict on this

Sometimes the argument makes itself, and this is one of those times.

Luca Guadagnino directed Artificial, starring Andrew Garfield as Sam Altman, dramatising the November 2023 weekend OpenAI’s board fired and then rehired him inside four days. Amazon MGM Studios developed the film, then dropped it from its slate in June 2026, a matter of months after Amazon committed 50 billion dollars to an expanded AWS and OpenAI partnership.

Amazon says the subject matter had nothing to do with the decision.

Netflix, Warner Bros, Focus Features and A24 all screened the finished film afterwards and passed. A24’s pass is the more interesting fact of the two.

The studio is backed by Thrive Capital, which holds a board seat at OpenAI and is one of its largest investors. Trade press cited political concerns more broadly across the other passes too.

Neon eventually picked the film up for a qualifying Oscar run. The film reportedly too hot to touch a few weeks earlier is now an awards season talking point, and Variety has already compared it to The Apprentice, the last time a film about a powerful man’s rise got stuck in distribution limbo for reasons nobody would quite put on the record.

Guadagnino himself, asked directly about the drop, wouldn’t discuss the deal. He did describe the rise of “this small oligarchy that wields truly radical control” when talking more broadly about AI’s effect on the world.

Nothing manufactures awards buzz quite like the appearance of trying to make a film disappear.

Whatever actually happened in that boardroom, a story where the industry’s biggest financial backer appears to wave a film about its own chief executive away from a release date does more damage to that executive’s reputation than the film would ever have managed on its own. Studios don’t usually need to be told twice which subjects are expensive to touch, and the fact that Amazon’s own name sits at the centre of this one is not a detail anyone in Hollywood has missed.

This is the small oligarchy argument from the section above, playing out in a second industry within a year of the first.

Build your own AI 2036

Everything above comes down to three open questions, not one. Compute, regulation and ownership each break a different way, and which way they break decides which future actually shows up.

Rather than tell you the answer again, this section asks you the questions this piece has been building toward, and shows you the ending that follows from how you answer them.

Question one, compute: “Does AI infrastructure keep pace with demand through the early 2030s?” Option A, grid connections and transformer supply catch up faster than the current build rate suggests. Option B, the current bottleneck holds through most of the decade, as the compute section of this piece argues is more likely.

Question two, regulation: “Does the direction Burnham’s government started in July 2026 hold?” Option A, binding rules, worker protections and audit trail requirements get enacted into law, in line with sizrok’s own forecast. Option B, the sandbox model survives largely intact under a different name.

Question three, ownership: “When AI’s productivity gains actually show up, who captures them?” Option A, gains spread via worker equity, sovereign compute and broad adoption across small businesses, the kind sizrok builds for. Option B, gains concentrate among whoever owns the models and the compute.

Outcome logic, eight combinations collapsing to four written outcomes, each built entirely from language already used earlier in the piece so the payoff feels earned rather than bolted on:

All A answers renders “Sizrok’s actual forecast”, the outcome this piece argues is most likely, referencing the worker equity, sovereign compute and Burnham material directly.

Compute A, regulation B, ownership B renders “The one we’re arguing against”, capacity arrives quickly but nobody built the guardrails first, referencing the concentration language from the nonsense section.

Compute B, regulation A, ownership A renders “Slow and steady”, the brake buys time for regulation and ownership structures to catch up before the infrastructure does, referencing the compute ceiling’s pace argument.

All B answers renders “Business as usual, slower”, the cautious default if nothing this piece expects actually happens, referencing the regulation section’s own honesty about the risk of stalling.

Each outcome screen ends with the same single CTA, “Book a conversation”, no separate CTA per outcome. Full build notes and exact outcome copy in the project brief.]

Frequently asked questions

What is sizrok’s headline prediction for AI in the UK by 2036?

Three things. AI reshapes jobs rather than deleting most of them, but not having basic AI skills becomes as limiting as not knowing how to use a computer used to be. The UK ends up back inside the EU in substance, driven by economics rather than sentiment. And the industry’s more dystopian predictions don’t hold, because compute and regulation both move slower than the loudest voices claim.

Will AI take my job by 2036?

Some tasks will be automated, particularly repetitive, single step, low judgement work. Most jobs get reshaped rather than removed, with the routine parts automated and the judgement, relationship and exception handling parts staying human. The real risk isn’t replacement, it’s not having the AI literacy that becomes a baseline expectation the way computer skills already are.

What is stopping AI from replacing more jobs faster?

Physical infrastructure, not model capability, at least for now. Roughly a third to half of planned 2026 US data centre capacity is delayed or cancelled, and the binding constraint is power grids and transformers rather than chips. That eases over the next decade as the current infrastructure wave finishes construction, at which point regulation and ownership structures become the real constraint, not compute.

Will the UK rejoin the EU?

That’s sizrok’s actual prediction, yes, well before 2036, even though no government has said so yet. The current government’s stated red lines rule out the single market, customs union and freedom of movement, but those are political positions, not laws, and the economics of staying outside keep getting harder to justify. We expect services trade, ecommerce and business relocation to be the practical drivers, with the political label catching up after the fact.

Who is the UK prime minister as of mid 2026?

Andy Burnham, sworn in on 20 July 2026 after Keir Starmer’s resignation the previous month. Burnham was mayor of Greater Manchester for nearly a decade before that, and has called publicly for tighter regulation of AI and Big Tech, with advisers briefing plans for British owned AI infrastructure and stronger worker protections against AI displacement.

Why is Manchester becoming an AI hub?

A combination of a strong university research base, a lower cost of operation than London, and a decade of deliberate regional investment. Greater Manchester’s AI sector is valued at 4.7 billion dollars in 2026 and the city has topped the UK’s AI readiness index outside London for three years running. Its former mayor is now the prime minister setting national AI policy.

Do I need to learn AI skills even if it doesn’t replace my job?

Yes, and it’s worth thinking about it the way computer literacy worked rather than the way job losses work. Nobody was fired in the early 2000s for being unsure around a computer, but within a decade it became a baseline expectation for almost every role. AI is running the same curve faster.

Is universal high income realistic by 2036?

It’s the proposal favoured by some of the loudest voices in AI for what happens once most jobs become optional. The bigger problem isn’t the mechanism, it’s who’s proposing it and why. Left to whoever owns the models and the compute, the more likely outcome without regulation is concentration, not universal anything, which is exactly what sizrok expects UK and EU regulation to prevent.

What jobs will AI create?

Roles that didn’t exist five years ago: people who scope and build automations, people who maintain and audit them once they’re live, and people who sell and support them with real technical fluency. All three scale with adoption, which is still low across UK SMEs, meaning the job creation side of this is still mostly ahead of us rather than behind us.

Will every business need a Chief AI Officer?

Most won’t need a full time one. IBM’s 2026 data shows 76% of large organisations now have a Chief AI Officer, up from 26% a year earlier, but for SMEs the emerging model is fractional, someone covering AI strategy for eight to fifteen hours a week rather than a six figure full time hire. Treat a full time hire with no budget behind it as a warning sign, not a strategy.

What has Elon Musk predicted about AI by 2036?

He’s told The Economist he expects AI to exceed human intelligence by around 2031 and for humans to lose control of its direction by 2036, with a 10 to 20% chance of a genuinely catastrophic outcome. Sizrok’s view is that his forecast on the technology might be right, but his preference for industry self regulation over government oversight shouldn’t be trusted given how much he personally profits from the outcome.

Why did Amazon drop Luca Guadagnino’s Sam Altman film?

Amazon says the subject matter had nothing to do with it. The timing, months after Amazon’s 50 billion dollar OpenAI investment, and A24’s own subsequent pass despite its backer holding an OpenAI board seat, has led most of the trade press to read the decision otherwise.

Book a conversation

None of this is a reason to wait and see what happens. It’s a reason to build the parts of it you can control now, before the businesses that moved first are the ones setting the terms for everyone else.

Book a conversation