Government Digital Service leaders Kalbir Sohi, Sonia Patel and Christine Bellamy were among the speakers sharing views at a recent public sector tech gathering
Asked to provide two words to characterise his feelings towards the current public sector artificial intelligence landscape, Kalbir Sohi is very quick to decide on the first of the two.
Six months into his tenure as the UK’s first government chief AI officer – a role which sits within the Government Digital Service – Sohi says that “one of my words is 'overwhelmed', because of the number of [different] things that people are talking about, the number of things that come along, [and] the 100 invites I get in my in my inbox every day”.
Speaking at the recent techUK Building the Smarter State conference in London, the second word chosen by the AI leader is a little more abstract: “chasm”.
“There’s a bit of a chasm between our ambition and where we are – and how we’re trying to get there,” Sohi explains. “I think it’s worth… staring that in the face and being relatively open about [the fact that] there’s a long way to go to achieve our ambition.”
Of course, a large part of the chief AI officer (CAIO) brief is to help bridge this abyss. There are many challenges in doing so and, according to Sohi, chief among them is “scale – which is probably the word that comes up in most conversations that I’m engaged in”.
“[People are] seeing these pockets of something that worked, but then [asking]: how do you share that? We have the problem in central government of moving sideways across departments,” he says. “The things that are working from my perspective are those where the idea of scale is built in from the beginning… if you set out and you’re a tiny team with a tiny capability, you could take on a tiny problem – and you might do a really good job on that problem. But then you might approach this question of: how do we scale that? And you don’t really have any of the mechanisms to go from that initial point to the next one.”
The CAIO tells attendees that the government agencies enjoying the most success in adopting AI are those that already have such mechanisms in place, and “can treat AI as an evolution of existing ways that they build and deliver services – with the same discovery methods and then the same scale methods”.
Risky business
Sohi arrived in government having previously worked for Spotify and Meta – organisations that he acknowledges have “a relatively high risk appetite” when compared with Whitehall departments. But there are still important lessons government can learn from the tech heavyweights’ approach to risk, he adds.
“One of the things I would say that both of those organisations were good at… is articulating where they are taking risks and where they are not taking risks,” Sohi says. “Something that I have been advocating a lot for in [GDS] – and then trying to take out across government – is that there will need to be some things where we can safely take greater risk, and there will be some things where we cannot; but we should be transparent [about that], and go in with our eyes open, as much as possible, to the reality of that situation.”
“There will need to be some things where we can safely take greater risk, and there will be some things where we cannot; but we should be transparent about that” Kalbir Sohi
In comments prefaced by an acknowledgement that “I struggle to sit on a panel and not say something controversial”, the AI leader also suggests that some of government’s methods – and the way they are led from the digital centre of GDS and its parent body, the Department for Science, Innovation and Technology – could stand to be a little more flexible.
“Right now there is a large amount of change,” he says. “And one of the things I [noticed] coming in is that we published an AI Playbook [in February 2025]. It’s really long, and it’s got loads of stuff in it – and a lot of that stuff is out of date now. So, I’m getting a relatively high amount of pressure to publish a new version. And I’m like: ‘Well, we’ll do a whole load of work and we’ll publish it – and then it’ll be wrong again.’
“Actually, in this faster-moving space – where we do have bigger opportunity and bigger ambition, but also much greater speed – I think we need a different set of structures that are a bit more adaptive.”
Sohi suggests that, rather than determining best practice to be followed across government, GDS should be highlighting and perpetuating the best of what is already happening in departments.
“We’re not necessarily [currently] incentivised to do this, but I actually would love for our central guidance to be whatever the Home Office found to work, [for example], rather than us trying to re-engineer [the guidance]. I would like us to get the stuff that’s blooming already to be the exemplar of good. I just feel like we’re going to be behind trying to centralise all the time, and it’s about [how] do we have some sort of adaptive system that pushes that information out.”
Enterprising work
An even more recent arrival at GDS is new interim government chief technology officer Sonia Patel, who joined in April having previously held the same post at NHS England. She replaces David Knott, who left at the end of 2025, and has been appointed to the post on a one-year contract.
Also appearing at the techUK event, the new CTO says that “GDS and DSIT have a critical role to play in convening and shaping cross-government thinking to help shape common standards, shared platforms, [and] architectural patterns”.
But, echoing Sohi’s sentiments on how centralisation can sometimes stymie progress, the CTO stresses that federated, rather than consolidated IT systems are the way forward – and that the UK could learn from various overseas counterparts.
“Countries like Estonia have shown how clear architectural backbone – not centralisation, but interoperability – can allow many systems to act as one government” Sonia Patel
“We need to learn from other nations in this journey towards an enterprise view of the state… [and] countries like Estonia have shown how clear architectural backbone – not centralisation, but interoperability – can allow many systems to act as one government,” Patel says. “In the US [as well], enterprise architecture is not an afterthought: it is a core discipline for reducing duplication, modernisation, driving and aligning investment, and delivering more coherent services. Singapore has also taken a whole-government approach, using enterprise architecture as a common blueprint to align systems, data and, more importantly, services. In Australia, architecture is used as a practical tool to guide investment, improve reuse and accelerate delivery.”
The tech chief characterises the infrastructure approach she wants to take as supporting part of a wider move from “digital government to intelligent government”.
“As we were just getting to grips with cloud, DevOps and product mindset, we’ve now got the opportunities that AI brings to us – because AI will not succeed in government if it’s simply layered onto the fragmented systems, inconsistent data and ageing infrastructure in front of us,” she adds.
There are currently many examples throughout the public sector “where innovation has been stalled or stopped because the foundation is too weak and fragmented”.
Bolstering these foundations, and ensuring that government is ready for the opportunities of artificial intelligence, will require “converging towards a modern enterprise architecture – and not just architecture that is a technical exercise, but an architecture as a strategic capability, one that helps us as government and public sector connect policy, services, operations, platforms and data into a coherent whole”, Patel says.
“We now have an opportunity to develop a shared view of modernisation: a view that aligns technology investment with priority outcomes for citizens and businesses, that enables interoperability rather than duplication, and creates the foundation for trusted, scalable AI adoption,” she adds “This is not about centralising everything: it’s about better equipping the system to deliver improved outcomes for citizens and businesses, delivering a modern intelligent government. Departments have different missions, different operating contexts and different needs; but there are areas of government where we can, and should, move together, and more deliberately, as one.”
Fronting up
A more familiar face from GDS – Christine Bellamy, formerly chief executive, and now interim director general for digital products – tells the event that, throughout the organisation’s near-15 years in operation, “what we’ve done really well is digitise the front end” of citizen services.
She adds: “We make it easier to for you to complete an application through great design patterns. We make it easier for you to find the service that you want through great search engines. And we grab hold of you and we take you to the place that you need to get to. But it’s a very siloed journey with lots and lots of dead ends.”
The issue facing GDS is “how to find the sweet spot between maintaining trust in services that people love – and they do love GOV.UK… it’s really well designed, and they’re able to get what they need – but [also] build on that brand, and take it forward”.
This progress is being conceived of in three main ways, the first of which is to “make the channels more straightforward, and find people where they are” – in a world where “95% of the population have got a phone”.
“People expect to log into it with their face, and they expect to get what they need from government and to do some quick curation and personalisation” Christine Bellamy
“They expect to log into it with their face, and they expect to get what they need from government and do that quickly and start to do some quick curation and personalisation,” she adds.
Secondly, GDS is thinking about how it uses chat interfaces to enhance user experience, Bellamy explains. A few days after her presentation, the GOV.UK Chat tool is fully incorporated into the GOV.UK App – but the products chief tells attendees that “I don’t think even that will make the big step change, although it may take away some friction”.
The third area cited by Bellamy – which is one she says is not without danger, but that “all the testing tells us that this is a bet worth exploring” – is greater use of AI agents.
“This is where we’ve got the biggest work to do – but it has probably got the most potential,” she says. “We know that people are using agents in their everyday lives – whether they know they are or not: this technology is just there in the products and services that they use. So, what does government do? Does it opt out? It could, but that’s dangerous. Does it opt in? In which case, why? And the key is that it has got to decide where it places its bets, because we haven’t got the most money in the world, and we’ve got to get that right.”
Bellamy adds: “I think the sweet spot is… where we can start to think about enhancing trust, maintaining the experience, but taking away friction, and seeing the individual as one person.”
It will be important to fully understand the “need states” in which people currently interact with government services, she says.
“We’re starting to spend time now thinking about real user cases that are complex, important, everyday-life need states for people in the UK – and they are problems that GOV.UK or individual departments find really hard to solve, and are probably also problems that you wouldn’t outsource to a third party,” Bellamy says. “In a world of big proliferation [of information and sources] there could be – although this is to be proved – some safety in bringing them back home to the world of GOV.UK, and saying: ‘If you come back here, it’s more likely to be trusted. We likely know who you are. We can make a connection to other agents. We can make connections to your banks. And we’re going to do it in a way that is explained, orderly, and where you feel in charge’.”
This article first appeared in the summer 2026 issue of Civil Service World. Read the summer issue online here