Why Building New Institutions Cannot Save Whitehall Alone
ARIA is brilliant. The AI Safety Institute proved that world-class state capacity is possible at pace. The Sovereign AI Unit is a genuinely exciting proposition. I am glad these institutions exist.
But if we are honest about what they represent, they are workarounds. They are concessions that the British state, as currently constituted, cannot do what we need it to do. And while we celebrate the workaround, we are avoiding the harder question: why can’t Whitehall do this itself?
The progress-oriented instinct is to build new. It is faster, cleaner, and more politically rewarding than wrestling with decades of accumulated dysfunction. You get to hire the people you want, set the culture from scratch, and sidestep the procurement nightmares and grade structures that suffocate ambition. But every new institution built to bypass the existing state is a tacit admission that we have given up on reforming it. And the existing state is where most of government happens, where public services are delivered, regulation is enforced, and the vast majority of public money is spent.
In ten years, will we look back and say that creating a handful of exceptional new bodies was sufficient to reform and transform the British state for the next century? Or will we find that we have created a two-tier system, where a few elite institutions are doing interesting work alongside a sprawling, unreformed civil service that still cannot translate political intent into public reality?
A Trap of New Institutions?
Despite political rhetoric often signalling the opposite, the number of central government public bodies grew from 474 in January 2015 to 603 in January 20251. The coalition government’s 2010 “bonfire of the quangos” cut the number of public bodies by more than a third2 — and yet they have since grown back and beyond. Not only has the number of these institutions increased over time, but often their remit and responsibility, and as a result, headcount, have too.
There is a real risk that new institutions become the very thing they were designed to escape: quangos with accumulated staff, diluted missions, and institutional self-preservation instincts indistinguishable from the bodies they were meant to replace.
The British Business Bank is an instructive example. It was established in 2013 with the purpose of supporting access to finance for businesses in Britain. Over time, it has become a one-stop shop for numerous categories of market failure. It has become an everything-institution. Its mission has bloated, and its focus has diffused, and the startups it is meant to serve are left navigating an organisation that is not quite sure what it wants to be.
The Sovereign AI Unit could be seen as a welcome counterpoint to a new institution, thankfully with a much more focused founding mission. It of course, does not have the exact same intentions as the BBB; it aims to support AI sovereignty by making strategic, concentrated bets on AI companies and is the kind of bold, commercially literate thinking we need. But it could become a step on the toes of everything institutions like the BBB, resulting in; it territorial confusion, duplicated effort, and wasted public money. The Sovereign AI Unit must be insulated from the rest of the government. It cannot afford to look like another unfocused state vehicle. It needs to be stacked with people who understand markets, technology, and risk, people who are emphatically not government bureaucrats. But what cannot be allowed to happen is that institutions like Sovereign AI become everything places of sprawling responsibilities, where ministers do not clearly tell the institution what not to do. It cannot be left responsible to fix market failures, which are international and out of the control of the government itself; what it can do is make good calculated bets under strong leadership from figures like James Wise. Therefore, the answer cannot always be to build new institutions to compensate for the failings of old ones. It must also be to fix the old ones, or scrap them entirely.
How can Whitehall successfully change?
So if we cannot rely solely on the creation of new institutions, what issues really matter for effective civil service reform?
Firstly, Whitehall cannot recruit or retain the people it needs. The civil service pay scales, grade structures, and hiring processes are designed for a world that no longer exists. The result is a talent deficit that new institutions temporarily solve by operating outside these constraints — but which persists everywhere else. The civil service Fast Stream is a case in point. In the 2024-25 application cycle, it received 72,691 applications for just 754 places — a 1% success rate, the lowest on record. With that volume of talent to choose from, the system should be selecting the very best. Instead, its hiring process relies on situational judgement tests that current Fast Streamers themselves describe as arbitrary, where passage feels down to luck as much as ability. I cannot emphasise enough how crackers these tests are. As a candidate, you have to sit at your laptop and are digitally placed in scenarios played out to you on video. For example, this may include being asked how to respond to a workplace incident of inappropriate behaviour – it is just as absurd as it sounds.
There is a broader point here about elitism. The civil service has historically treated the idea of recruiting from elite universities as contrary to social equality. It is valid for policymakers to want the civil service to be representative. But too many policy interventions cancel each other out. We have successfully invested time and energy in making Oxford and Cambridge, and other elite institutions, look a lot more like the country they serve by improving the talent pipeline further upstream. By the time someone is applying for a graduate programme, the civil service should be selecting unapologetically for capability. Other interventions exist for other purposes. The Fast Stream’s primary motivation should not be rectifying structural inequalities; rather, it should be hiring the most talented people to run the British state.
Therefore, to create new institutions which are exempt from the civil service hiring processes is a good thing, but this should not blind us from the existing failures in the system for hiring that remain across the majority of Whitehall.
The Regulatory State
Nowhere is the need for reform more urgent than in regulation. The instinct of progress-oriented people is often to work around regulators, to seek exemptions, sandboxes, and carve-outs. But the fundamental question of whether our existing regulators are fit for purpose keeps getting deferred.
The FCA’s regulatory sandbox is a key example. Launched in 2016, it was genuinely revolutionary — arguably the most impressive regulatory innovation any British institution has produced in the last decade. The outcomes have been strong: firms entering the sandbox saw a 15% increase in capital raised3. The sandbox has been exported internationally, too, to many other jurisdictions as best practice.
But the sandbox has not transformed the FCA. It has been a bolt-on, a brilliant bolt-on, but a bolt-on nonetheless. The FCA itself remains largely unchanged. This is the pattern: we create innovative new features within existing institutions, declare success, and move on, leaving the underlying organisation unreformed.
Consider the current debate around social media restrictions for children. This is, at its core, a question about whether existing regulators, Ofcom in particular, are equipped to handle decisions that are fundamentally political in nature. Regulating social media is not a technical exercise in market design. It involves value judgements about childhood, attention, mental health, and the boundaries of corporate responsibility. Are our regulators capable of making those judgements? Should they be the ones making them? Or have we outsourced political questions to technocratic bodies that were never designed to answer them?
As a result, even if we are not willing to scrap regulators, bringing back meaningful ministerial responsibility to them will be critical. The CMA’s recent interventions on AI partnerships and mergers are a case in point. Between December 2023 and early 2025, the CMA opened five separate merger investigations into AI partnerships — Microsoft/OpenAI, Microsoft/Mistral, Microsoft/Inflection, Amazon/Anthropic, and Google/Anthropic4. The result of these investigations? All were either cleared or found not to constitute relevant merger situations. The Microsoft/OpenAI investigation alone consumed 15 months of pre-notification investigation, only for the CMA to dismiss the case. This has to be an area where ministers must intervene to set the strategic framework for intervention and conduct, and then day-to-day operations are run by the regulator.
Scrap What Does Not Work
The abolition of NHS England, the single largest arms-length body in government, offers a precedent. Scrapping a major public body is difficult and messy. But the decision reflects a recognition that institutional architecture matters, and that sometimes the structure itself is the problem.
Therefore, if we are serious about transforming the British state, we need to stop treating institutional creation as a substitute for institutional reform. Both are necessary. But the hard, unglamorous, politically costly work of reforming existing structures is the part we keep avoiding. The cumulative effect of improper reform is a state that grows in complexity without growing in capability. Reducing headcount, as the government has pledged, is critical, but headcount currently reflects the accumulated burden of statutory duties, including replying to parliamentarians’ letters, processing FOI requests, and preparing for judicial review. Cutting staff without reforming the underlying obligations merely guarantees worse service delivery.
This means that we must be willing to scrap regulators that are not fit for purpose. It means confronting the civil service’s structural inability to attract and retain top talent across the board — not just in bespoke new units. It means devolving power and resources to local government rather than hoarding them in Whitehall. It means accepting that the trade-offs and write-offs involved in genuine reform are the price of a state that works.
New institutions are a bet on the future. But if we are not simultaneously making the painful, necessary reforms to the institutions we already have, we are placing that bet on a foundation that is already cracking. And no amount of new buildings will save a state whose old ones are collapsing around it.
Meanwhile, the failure to transform public services to meet rising expectations creates visible failures and rising costs without improved outcomes. Citizens now benchmark government services against consumer experiences, like next-day delivery, instant customer service, AI-driven personalisation, and the gap between what the private sector delivers and what the state provides grows wider every year. Trust in the state’s competence erodes further.
The government remains unwilling to make the decisions necessary to change how it structures itself. Reorganisation is politically expensive and bureaucratically resisted. It is easier to announce a new unit than to abolish an old one, and easier to add a function than to remove one.
Excellent piece - true ministerial responsibility with focused targets and priorities over departments has to be the starting point
Very good