No ivory towers here please
Departments come and go, but habits can last for decades. Whether government builds or buys, owns or outsources, or unlocks the best of both, those habits are much harder to change.
That’s worth remembering as Whitehall absorbs the news that the Prime Minister has abolished the Department for Science, Innovation, and Technology, redistributing its functions across government. It is at least the third major shape this work has taken since 2007, and the third Secretary of State in three years for many of the people doing it.
Another reorg. That’s the first reaction inside Whitehall, and it’s not an unreasonable one.
There’s a palpable fatigue, and it deserves more than a passing shrug and nod of sympathy. Nobody does their best work while waiting to find out which building they’ll be reporting to next month, and the teams inside DSIT, many of which are excellent, have earned the right to be tired of hearing “this time it’s different.”
If any review of the DSIT disbanding should carry a starting assumption, it’s that the people, not the department, are the asset worth protecting.
But that does not mean structure is irrelevant. Structures determine ownership, incentives, and accountability. They shape what government is able to do. Just glance at Conway’s Law and this reveals itself. The mistake is believing that a structural change alone is a strategy.
Sometimes the org chart becomes an alibi.
There’s a reasonable case that devolution, the project Andy Burnham has spent a career on, needs digital and AI capability working from shared foundations, so that no department, council, or regional government continues to reinvent the wheel with its own tech stack.
But shared ownership alone does not create shared capability. The fact that digital and AI have sat together to date has not, by itself, limited sprawl. If anything, tech sprawl has worsened.
Nonetheless, splitting policy from delivery risks losing exactly the muscle that agenda depends on. This argument should be taken incredibly seriously, even if the high level departmental split has already happened. Whichever way the org chart falls, it isn’t actually the thing that will decide whether devolution succeeds.
Which is precisely why one thing worth protecting regardless of outcome is Dame Chi Onwurah‘s Science, Innovation, and Technology Committee – a function that has spent this Parliament asking the questions nobody else dared to. The committee has nothing to do with which logo sits on the building, and it should survive this reshuffle even though the department hasn’t.
Beyond the committee, there is plenty of credit due to the rest of DSIT. GOV.UK remains one of the best citizen-facing digital services anywhere in government, and arguably anywhere in the world. GOV.UK Notify and GOV.UK Pay are extending that legacy, and have the ingredients to become genuinely open source if they welcome community contribution. Nobody serious should pretend the last decade of work here was wasted.
What DSIT hasn’t done is build sovereign and resilient digital infrastructure below that GOV.UK front door. Standardising the citizen-facing layer is a different achievement from owning the plumbing underneath it, and for years now too much of that plumbing has come to depend on two American hyperscale cloud ecosystems. “Cloud first” was only ever meant to mean buying commodity infrastructure instead of building your own, but it became full-stack dependency instead. “Open source first” became code-in-the-open when there’s time for it, which isn’t the same policy at all.
That gap is what a reset should actually close, wherever GDS and the Incubator for Artificial Intelligence end up sitting.
Whatever the intention behind it, this reshuffle creates a rare moment when decades of inherited assumptions can be challenged instead of merely adopted. A few principles would turn it from a redrawn org chart into a rebuilt operating model, and would give Kanishka Narayan MP, deservingly promoted to the Cabinet table as Minister for AI, something more useful to inherit than a fresh set of initials.
Build together, not just in the open. Public money should produce public code. Nothing new there, but it should be a working method, not a slogan for a press release. Government should build upstream first, then package stable, supported downstream releases for live public services – much as Red Hat has done for years with Fedora and RHEL. Today, you can inspect GOV.UK Forms, Pay, and Notify. In most cases, you still cannot meaningfully contribute improvements upstream. Small central teams can’t keep pace with demand, so departments spin up their own services instead. Even new non-GOV.UK sites are appearing to nobody’s delight. InnerSourcing should be the baseline default, and true Open Sourcing the expectation, with genuine exceptions signed off by GDS itself (or whatever it evolves into), not nodded through by a deputy director on a Friday afternoon.
Buy commodities, not permanent tenancy. Every machinery-of-government change reshapes how teams build things underneath it, whether anyone plans that or not, which makes this exactly the moment to work out what’s owned where, and how much is duplicated across departments. Until now, nobody truly had the mandate to ask. We need Wardley Maps, Team Topologies, Domain Driven Design, and Value Stream Maps to highlight opportunities to commoditise. The Technology Code of Practice should enable commoditisation, but it needs rewriting so that concepts such as “cloud first” and “open source first” describe what departments must do, not what they were once encouraged to consider. A shared, FinOps-style cross-charging function, helping any department to productise a service and offer it to another without everything having to live inside GDS, would turn the centre from a bottleneck into a marketplace.
Keep the exits open. Who should be the person signing off the decision to process British data through a foreign data centre, against NCSC guidance and secure-by-design principles? Surely that call belongs somewhere pretty far up the ladder? There are countless similarly disconcerting decisions, especially related to proprietary dependency, being made across departments today. Government needs a standing team whose job is finding lock-in related risk before it hardens into permanence, and helping departments out of it before the exit cost becomes politically unpayable. That’s the same commercial and procurement muscle that was already being built inside DSIT and now badly needs protecting and elevating, not scattering, through the move.
Make British data build British value. At present, the UK public sector is providing fuel for other nations’ AI economies; supplying the data, absorbing the disruption, paying the license fee, while long-term value accrues elsewhere. The alternative involves shared standards, sovereign compute, open and interoperable models, and a functioning market for British data. It starts with one procurement principle. No major public sector AI contract should be signed unless the solution is portable, the systems are interoperable, and there’s a credible, costed route to switch provider. This isn’t purely a preference for British-flagged vendors over global ones. It’s a preference for openness, including enterprise open source, wherever it can be found. It’s a refusal to sign away leverage before anyone’s worked out what it’s worth.
None of this work needs DSIT’s name on the door. But it does need GDS and the Incubator for Artificial Intelligenceto treat this change as a mandate to serve, support, and enable the rest of government, rather than a refurbishment on a tower to rule from.
That’s the opportunity hiding inside a reorganisation nobody asked for. No ivory towers. Just a government that remembers how to build world leading capability.
Originally posted here
