The lesson from the UK: classify AI as a national core development project while cutting the Ministry of Science and Technology

UK’s new Prime Minister Andy Burnham fired the UK’s Department for Science, Innovation and Technology (DSIT) the day after taking office—despite it having been set up only two years and five months earlier by former PM Rishi Sunak. Yet in the same week, he appointed 36-year-old Kanishka Narayan as the UK’s first-ever AI minister to sit in Cabinet. Also in the same week, the National Audit Office released a report that directly called out a lack of consistent estimation methods behind the government’s much-touted claim that “digital transformation and AI can save 45 billion pounds a year.” (Background: The UK FCA warns about a “AI arms race” in finance—urging regulators to rapidly regulate big models like ChatGPT, Claude, Gemini, etc.) (Background note: Taiwan’s government is actively planning “sovereign AI”; National Science and Technology Council chairperson Wu Cheng-wen: forming a National Artificial Intelligence Strategy Special Committee)

Table of contents

Toggle

• The UK that week
• The UK Cabinet’s special kind of weird
• No denominators
• Taiwan, too

In the 1980s BBC political satire Yes, Prime Minister, Deputy Permanent Secretary Humphrey Appleby has a set of “politician’s syllogism” that has been quoted for four decades. Step one: we have to do something. Step two: this is what it is. Step three: therefore we must do it. The reason this line is still circulating today is not because it’s funny—it’s because it’s too hard to get wrong.

From July 20 to 21, 2026, the UK government played out this whole syllogism in less than 48 hours.

On July 20, after meeting King Charles III at Buckingham Palace, Andy Burnham returned to the gates of 10 Downing Street to swear to end political turmoil. He became the UK’s 59th prime minister and the seventh within a decade since the 2016 Brexit referendum. The next day, he dissolved the UK’s only central department named after technology. The next day, he also put AI—this was the first time—into the Cabinet meeting table.

Most media covered these two events separately: one was bureaucratic reorganization, and one was personnel appointments.

The UK that week

First, we need to explain the UK political situation, because Taiwanese readers are generally not familiar with the background of this round of leadership changes in the UK.

Keir Starmer led Labour to a landslide victory in July 2024, but within two years his personal approval collapsed. Polling showed voters believed he had brought no visible changes. On June 22, 2026, with Labour MPs broadly losing confidence internally, Starmer announced he was stepping down as party leader and prime minister.

The problem was that Labour’s party constitution stipulates that only members of Parliament can run for party leadership—yet at the time, the loudest contender, Andy Burnham, was the mayor of Greater Manchester, not an MP. The solution was a by-election. After an MP resigned his seat in Makerfield, Burnham won the by-election on June 18. Then on July 17, with nominations from 379 MPs (more than 94% of Labour’s total MPs), he was endorsed as party leader with no competing opponents. On July 20, he officially took office as prime minister.

On the second day after taking office, he moved the org chart at Whitehall.

What was shut down was the Department for Science, Innovation and Technology, established on February 7, 2023 in a reshuffle under then-prime minister Rishi Sunak. The approach was to split the Department for Business, Energy and Industrial Strategy into three parts, then take the digital-policy remit from the Department for Digital, Culture, Media and Sport and stitch it into a new department. The first minister was Michelle Donelan, and it lasted two years and five months.

Its functions were split into three parts. Technology and science policy were merged into the newly formed Department for Business, Innovation, Science and Trade (DBIST), run by Jonathan Reynolds. AI policy and the inflow of AI into the public sector were moved to the Cabinet Office. Digital government functions—along with the Government Digital Service (GDS) and the Central Digital and Data Office (CDDO)—went back to the renamed Department for Digital, Culture, Media and Sport. Former science minister Liz Kendall left office, and the “technology minister” post disappeared from the UK Cabinet.

On the same day, Kanishka Narayan was appointed as AI minister—spanning both the Cabinet Office and DBIST—and he sits in Cabinet. This is the first time AI has obtained a Cabinet-level seat in the UK government.

Narayan’s résumé is itself a story. Born in November 1989 in Muzaffarpur, Bihar, India, he moved with his family to Cardiff, Wales at age 12. The family lived in a one-bedroom apartment, and he walked to attend public school. Later he studied Politics, Philosophy and Economics at Eton College and Oxford University’s Balliol College, then earned an MBA at Stanford. Before entering Parliament, he worked as a civil servant in the Cabinet Office during David Cameron’s tenure and in the Department for Environment during Liz Truss’s tenure. In July 2024, he was elected as MP for the Vale of Glamorgan constituency, becoming the first MP in Welsh history with a minority ethnic background. In September 2025, he took over as Parliamentary Under-Secretary of State for AI and Online Safety, and within less than a year he was promoted to Cabinet.

When he took office, he said AI is probably the most important technology in human history, and what it can bring is a reindustrialized UK, stronger national security, and public services that are genuinely improved. He also acknowledged there are risks, and he acknowledged public anxiety about jobs and speed. There’s nothing particularly objectionable about what he said, and even that tone is very sincere. The problem is not what he said—it’s that he said it on the second day after the technology department was dissolved.

The UK Cabinet’s special kind of weird

To understand a government’s real stance on something, it’s better to look at what it shuts than what it opens.

Creating a ministry is a political statement. You need a reshuffle, an announcement, new org charts. Shutting a ministry is an organizational price tag. You have to cut into existing budgets, existing personnel, and existing statutory positions—and you have to be willing to take on the costs of doing it all over again.

The former tells you what the government wants to be seen as. The latter tells you how much this Cabinet is willing to pay for it.

The UK tech industry’s response was remarkably consistent. techUK and the Startup Coalition co-signed a letter to Burnham, describing the plan as “the wrong change at the wrong time.” Their argument was not about attachment to ministries; they said dismantling itself has a cost. In their letter, they specifically called out the AI Safety Institute, the Sovereign AI Fund, the GDS, and the UK Research and Innovation (UKRI) activities—arguing that breaking them up will slow everything down precisely when speed is needed most.

Dom Hallas of the Startup Coalition put it even more plainly. Putting tech inside an oversized business super-ministry is like making UK tech companies compete for the same minister’s attention as heavy industries like steel. Matt Clifford—who previously served as Keir Starmer’s AI opportunities adviser—called it “a big mistake.” Ian Hogarth, a venture capital partner and also chair of the UK AI Safety Institute, was the least emotional—and the sharpest. He said a reorganization at that scale would basically distract the relevant departments for a year.

The Burnham government’s argument is that only by putting business, science and trade together can it deliver a coordinated industrial strategy. That rationale isn’t far-fetched. DSIT was originally carved out of the business ministry; after being carved out, tech policy and industrial policy indeed went their separate ways.

But the two sides are talking a bit past each other. The government talks about policy coordination; industry talks about attention allocation. Coordination can be solved via process; attention cannot. A minister has only so many hours in a day. And once you put steel, trade negotiations, industrial subsidies, and AI under the same roof, AI shifts—from being a served area to a queued-up agenda item.

Narayan’s new Cabinet seat was designed to plug that gap. Being in Cabinet, in theory, lets him pull resources across departments. This is an upgrade. But the upgrade is one person; the downgrade is an entire system of staffing. People rotate; structures accumulate.

No denominators

What truly makes this week worth recording isn’t the org chart—it’s the appearance of two numbers in the same week.

The first number is £1.8 billion. This is what the UK Office for Budget Responsibility (OBR) estimated in November 2025 for the three-year cost of a nationwide digital identity card called the “Brit Card,” covering fiscal years 2026-27 through 2028-29. The card was originally meant to list name, nationality, date of birth, residency status, and a photo. Its purpose was to check right-to-work, thereby cracking down on illegal employment. It was announced in September 2025, immediately triggering the second-largest petition in UK parliamentary history, with nearly 3 million people signing in opposition. In October 2025, the government was forced to remove the mandatory requirement and switch to voluntary use, allowing existing documents like passports to be used as well. Throughout 2025, Keir Starmer defended it, even holding closed-door meetings to work out how to implement it. In July 2026, after Burnham took office, it was scrapped outright.

Where did the money go? It’s worth clarifying, because outsiders at one point thought it would be diverted to AI. In fact, Burnham’s government said the resource would be used to cover household electricity bills, as part of support to ease the cost of living. This is a political judgment, not a technology judgment.

The second number is £45 billion. This is the headline figure the UK government uses to describe how much money “digital transformation and AI can save each year.” It came from DSIT’s January 2025 report State of digital government review. The report said that if the public sector were fully digitized, there would be more than £45 billion a year in “unrealized savings and productivity benefits”—equivalent to 4% to 7% of public spending. This number was later repeatedly cited by the prime minister and ministers, becoming one of the foundations for fiscal arguments.

How was it calculated? DSIT’s assumption was that 100% of routine tasks and 10% of non-routine tasks could be automated.

“It’s a reasonable guess, but it’s obviously a guess. It’s very hard to measure.”
Laura Gilbert, former i.AI (government AI accelerator) director, said this when testifying in Parliament about the £45 billion.

Gilbert is not an enemy of this number. She spent more than four years at 10 Downing Street as founding director of 10DS—the government’s data science team—and i.AI. She is one of the best-informed people in the UK government about this. She spoke honestly, and ironically, that honesty is more unsettling here.

Nick Davies of the think tank Institute for Government put it more directly. He said that unless you have a serious plan to cut a substantial amount of manpower or reduce capital expenditure, reaching that £45 billion would be very difficult. He added that as of now, nobody has clearly defined what counts as “routine tasks,” and he even doubts whether the government itself knows.

In its report Rewiring the state, the House of Commons’ Science, Innovation and Technology Committee directly labeled this number as “worryingly optimistic,” and listed it as an example of “hype led by ministers,” arguing that this kind of hype doesn’t drive digital transformation—it drags it down.

Then came the National Audit Office. On July 15, NAO published Government workforce planning: Lessons learned, and used auditing language to say the same thing again. The report pointed out that the UK public sector employs 6.2 million people, with people costs of £260 billion. 76% of civil servants expected AI to change how they work within five years. But the efficiency plans published by each department “did not provide details on how the expected workforce efficiencies would be delivered,” and there was no consistent methodology across government. Cabinet Secretary Antonia Romeo herself had said that the civil service’s workforce plans “do not sufficiently reflect how AI and technology will change what the civil service will look like.” NAO also delivered another blow: the public sector itself lacks the technical and digital talent needed to achieve these efficiencies.

Put these two numbers side by side, and you get the shape of this week.

£1.8 billion is spending. It has clear annual time ranges, backed by an independent body, included in budget documents, audited—and therefore it can be cut. £45 billion is savings. It lacks consistent methodology, lacks a verifiable denominator, comes from an assumption about “routine tasks,” and it hasn’t been cut so far because nobody is required to attach it to who should be accountable.

“The government cut the number that can be verified, and left the number that can’t.”

That’s what I think makes this week truly worth recording. In accounting, spending and savings are symmetrical—you cut one and add the other. In politics, they are completely asymmetrical. Spending must be defended; savings only need to be believed. When a fiscal gap is staring you in the face, a savings figure without a denominator is the most convenient filler. It doesn’t need a tender, doesn’t need progress reports, doesn’t need to manage delays, and doesn’t have to face nearly 3 million people’s petition.

And now AI is the easiest kind of carrier for such numbers. It’s new enough that there’s no historical benchmark to call it out. It’s broad enough that any efficiency can be attached to its name. And it’s professional enough that anyone who questions it can be dismissed as not understanding technology.

DSIT lasted two years and five months. From the announcement to scrapping, the Brit Card took 10 months. Labour took two years from a landslide win to replacing its prime minister. Since the Brexit referendum, the UK has replaced seven prime ministers.

Taiwan, too

Taiwan’s Ministry of Digital Affairs was established in August 2022. Today it’s nearly four years in, and debates about it have never stopped. The questions—whether specialized ministries should remain centralized functions or whether resources should be concentrated in the same ministry or embedded across ministries—are exactly the same issues Taiwan and the UK have been arguing about.

The UK provides a ready-made case, and even a negative-direction case. It shows that a ministry created just to “signal that it’s valued” may, in the eyes of the next prime minister (president), be nothing more than a puzzle piece that can be merged. It also shows that when you fold technology into a large ministry, what industry worries about first isn’t the policy direction—it’s whether they’ll even get onto the minister’s calendar.

I don’t think Taiwan should draw conclusions like “so we must keep the Ministry of Digital Affairs” or “so we should abolish it.” Both conclusions are too cheap. What really should be asked is: since its establishment, what has this ministry accumulated that others cannot take away? If the answer is a batch of real technical talent that can truly be retained, a data and identity infrastructure that can be reused across ministries, and a few completed projects that have been verified effective—then it’s not something that can be dismantled.

If the answer is mainly a set of new org charts and several annual policy plans, then it doesn’t differ from the governance problems the UK faces.

For politicians, “controversial” means it loses votes. “Courageous” means you lose an election.