Thoughts from the mind of Ben Welby

Tag: AI

Pocket, Pavement, Platform: Government in the App Store and on the High Street – Part 3

This was one big post, and now it’s five smaller pieces thinking about what public service really means in a digital age, and the risks of mistaking convenience for coherence. I started by wondering about how far fitting government into our pockets offers real transformation. In the last post, the topic was the underlying plumbing that makes everything else possible. The next piece will argue for an omnichannel approach that designs for every doorway. And when you make it to the end then your reward is a piece that is all about Goths.

But now, in this third part, I want to you to think about the future (which isn’t too far from being the present): where where the interface melts away altogether. What happens when services are no longer tapped, but summoned? As AI agents emerge, does that realise the dream of transformation, or is it just that it keeps complexity out of sight?

Disappearing interfaces don’t disappear the problem

If apps promise pocket government, AI now promises agentic government: services summoned through conversation, no forms or websites needed, just a natural interface that handles everything for you. It’s an appealing vision, and maybe not far off in some domains. But abstraction without foundation risks leaving people behind.

Apps, when done right, can be transformative. They can bring government closer, offering convenience and speed for those who want it. The GOV.UK App and the wider GOV.UK ecosystem could deliver that promise. Pick the most recent government service you interacted with and imagine its app-enabled future.

For me that’s renewing my driving licence: a push notification from the GOV.UK App, thumbprint authentication (GOV.UK One Login), reusing a passport photo (Home Office), paying via GOV.UK Pay, confirming via GOV.UK Notify and a renewed credential in my GOV.UK Wallet. A seamless journey in seconds, where the user barely notices the machinery – DVLA, Home Office, GDS, or otherwise – because the ecosystem just works. Apps shine for tasks like these – quick, personal, and always on hand; when the infrastructure supports them.

That’s also the transactional promise GOV.UK has offered since 2012: one platform, one ecosystem with one consistent user journey. And, in 2012 and still today, that vision demands simple, integrated, permissioned services: plumbing that works and data that flows. 

But without that plumbing then an app is just another channel, not a platform. Right now the GOV.UK App feels like it’s a solution in search of a problem. In being distinct from the GOV.UK website, for which 100% of government services are built it’s introducing friction – like requiring authentication to access a website that takes people to services using different ways to log in.

Back in 2013, GDS famously declared: “We’re not ‘appy. Not ‘appy at all.” The principle was clear: standalone apps must wait unless the core web service works as well on mobile, and even then, only by exception and driven by user need. Do not read my callback to that as an oversimplified holding to an outmoded point of view. A decade on, as user needs have evolved and so has technology, apps have a clear and valuable role.

But for government they should always be additive to the web experience. Digital inclusion is not a solved problem and while releasing early and failing fast has its merits, there is a deliberate decision to launch the GOV.UK app before the core web service meets that bar and with the open expectation that many features are going to be exclusive to the app creates a walled garden, not open doors. And for me that runs counter to what made GOV.UK a global exemplar in the first place.

AI amplifies this challenge. An AI-led bit of government in your pocket might navigate complex services but it can’t fix contradictory policies, confusing eligibility, or poor service design. I’ve learnt so much from my vibe coded experiments, one of which was to create an AI-led experience of jobs and careers support. But that example also clearly showed that the value lies not in the interface but in the underlying service.

Anthropomorphising AI is obviously not the right thing to do, but thinking about an AI agent like a person might be. It’s the work of service design – figuring out how to best help someone achieve an outcome. When you design for people whose interfaces onto the service might not be directly through a browser but indirectly via their children or a support worker then that delegated experience also reflects something of the experience for those whose interface of choice is AI.

Indeed, over time, some people will experience a disappearing interface. Entire service journeys will be handled by agents. But right now, no UK government service is designed with that in mind. They’re designed, and as long as the Service Manual and the Service Standard exist, will continue to be designed, to lower the barriers to entry and include everyone. They’re rightly not locked behind an app layer or forcing you to authenticate before you get to the content you need. That safeguards the state as service-shaped, interoperable, and testable, paving the way for an AI-mediated future without excluding anyone today.

Whether or not it’s what Martha Lane Fox had in mind, this is really the embodiment of what it means for government to go wholesale. After creating the digital centre, fixing publishing, and fixing services, the final task was to build the state as a platform: a network of capabilities, not a stack of destinations. Open APIs, shared infrastructure, and services that can flow into the places people already are. Useful then but now essential in this potentially agentic world of ours.

So AI definitely has a role. But it’s a layer, not a solution. A reflection of good service design, not a replacement for it. And any AI-led experience must be one of many. Because for all those who talk to bots, there are plenty who need a human to sit with them on the sofa over a cup of tea.

This is where transformed government shines: services designed for everyone. And that discipline must extend to every channel, digital or physical, to keep the state inclusive. A state built for everyone doesn’t retreat behind an app icon, or vanish into AI. It shows up: for real lives, in real time, across real channels.

You’ve made it past the half way point of my five part series. Next in Part 4, it’s all about the real world and exploring how public services must meet people where they are, not just through screens, but through every available entrance.

In Part 1 I interrogated the appeal of “government in your pocket” and whether it is more valuable than simply being a good soundbite. Part 2 went beneath that surface: to the plumbing that makes service delivery possible. And in the last part we’ll talk about putting GOV.UK on the High Street.

Pocket, Pavement, Platform: Government in the App Store and on the High Street – Part 2

This was one big post, and now it’s five smaller pieces thinking about what public service really means in a digital age, and the risks of mistaking convenience for coherence. In Part 1, I wondered about how far fitting government into our pockets offers real transformation, or just a sleeker surface. In the third the focus is on what does it mean for services to be completely AI-led. The fourth argues for an omnichannel approach that designs for every doorway. And the final piece is all about Goths.

But now in Part 2, I want to look beneath that surface: to the plumbing that makes service delivery possible. Because no matter how beautiful the interface, it’s only as strong as the data, infrastructure and coherence behind it.

Pipes before Pixels

Ukraine’s Diia app is the gold standard for what pocket government could look like and undoubtedly an inspiration for governments wanting to make a statement about transformation. Diia works because of years of relentless infrastructure work: national ID, open APIs, robust data registers and long-term political will, sharpened by the focusing energies of being at war. Diia isn’t magic. It’s a lot of unsung hard work paying off in a crisis. It’s the product of infrastructure work that too often gets ignored in glossy launches elsewhere.

Similarly, Portugal have been doing transformative things, enabled by underlying infrastructure, since before GOV.UK existed. Back in 2010 they used integrated data to automatically enrol eligible people (7% of the population!) to access a special tariff to reduce their energy costs. In the UK we’re relying on Nationwide, EntitledTo and Turn2Us, to support people to receive up to £23bn of benefits that otherwise remains unclaimed.

The UK has remarkable teams and a superb toolkit for building world-class services. But the pipes leak. Data is patchy, open government data has been abandoned1, and for all the enthusiasm about the latest approach to digital identity2 we remain out of step with our international peers. Services remain siloed, stitched together at the front-end but not the back. An app that forces OneLogin onto you in order to browse a website, through which you’ll still have to use other authentication methods isn’t solving this – it’s just adding friction. People without smartphones, or confidence, or connectivity, are pushed further out.

Focusing on the interface, whether that’s an app or a website, without fixing the underlying data flows or service design risks building on sand. A truly joined-up state requires shared data infrastructure that enables seamless services, whether accessed via a tap, a call, or a walk-in. Without it, digital promises remain just that: promises.

None of this is to deny the value of good apps. They do reduce friction and make things easier and more convenient – there are things you can’t do with the web that you can do more readily in something with persistent state and hooked into a wider, more personalised interface. But apps are only ever as good as the foundations beneath them, foundations that benefit all those who are not persuaded by the idea of installing an app for an occasional, ‘once and done’, need.

With strong plumbing, user-centred design, and joined-up infrastructure, an app can be a joy. But if that doesn’t exist, it’s just a Potemkin interface: a facade that crumbles when pressed.

This was quite a short piece in my five part series so read on to Part 3 to ask what happens when the facade disappears altogether. When AI agents replace screens, can they carry the weight of transformation, or just hide its absence?

In Part 1 I interrogated the appeal of “government in your pocket” and whether it is more valuable than simply being a good soundbite. Part 4 returns to one of my soapboxes in arguing for an omnichannel approach. And in the last part we’ll talk about putting GOV.UK on the High Street.

  1. The OECD’s OURdata Index helpfully tracks the ebbing away from being a leader to lagging behind in just six years. ↩︎
  2. Consider how the BlueSky community has got into a frenzy about identity verification due to some not brilliant legislation which leaves the platform having to scrabble around rather than the burden being carried by an effective, trusted and whole of society approach to proving that you are who you say you are have the attributes you claim to have on the Internet as well as in person.  ↩︎

Vibe Coding, Fireworks and the Mortar of Government

A few days ago, I lit the fuse on a working prototype of a government service. No team, no procurement cycle, no waiting for approval. Just me, a few prompts, and a handful of AI tools. And honestly? Fireworks.

Vibe coding (or vibecoding) is an approach to producing software by using artificial intelligence (AI), where a person describes a problem in a few sentences as a prompt to a large language model (LLM) tuned for coding. The LLM generates software based on the description, shifting the programmer’s role from manual coding to guiding, testing, and refining the AI-generated source code.

Vibe coding, Wikipedia

I’m not new to what’s now being called vibe coding. Over the last year ChatGPT has helped me to bring a few random ideas to life1. Last weekend I thought I’d see what Codex CLI could do and I was again blown away. I mentioned this at work and in the conversation that followed we mused on whether some of the frustrations we’d been feeling could be shifted by trying the same thing there.

So I sat down with a laptop, some product instinct, and a handful of different AI tools. I wanted to see whether we could finally conjure the ‘fireworks’ we’d been waiting weeks to set off. I started with ChatGPT and the scale of the task was a bit intimidating. But then I remembered about Firebase and in minutes had something to show off. As I did, another colleague responded by asking if I’d seen Stitch, and another colleague said I should check out Jules.

And once I discovered Jules, that was when things got really interesting. Very quickly I had something live. Not a sketch or simulation, but something real. It’s up and running on Render (and I’d love to give you the link but I probably shouldn’t let it escape into the wild; at least not yet).

Obviously it’s just a prototype. But that also seems to do it a disservice. What is true is that it absolutely appears to do the job we had in mind. No engineers. No designers. Just me, some prompts and decisions, and it works, and it works in a way that will absolutely elicit the right sort of oohs and aahs.

I suppose I ought to make one small confession. I really shouldn’t have done any of this. Inside the department, everything except Copilot is blocked (and even then you only get Copilot on a Windows machine, not a Mac). Which means this burst of delivery joy has happened off network, off platform, and probably against better judgment. But that, too, is part of the problem. When the path of least resistance leads outside the system, it’s the system that needs fixing, not the people finding their way around it. Well, I would say that wouldn’t I?

Now, for our purposes as a team this exercise might be the perfect fireworks but more broadly for government, what are the repercussions?

I’m going to call it: Jules and Codex earn their hype.

Continue reading

Back to work

When I left the OECD last summer, it was for several reasons (some of them implicit in this blog post). And I’m pleased that I decided to do so. This has been such a valuable slice of time – a luxury I recognise few people are able to enjoy.

I’m so lucky to have had the career I’ve had to this point and the breadth and depth of what I’ve done. But it has given me a bit of a dilemma about the shape of who I am and the role that best fits, especially in this new, post-OECD phase.

In terms of substance, the OECD was a good match and I was fortunate to have been adept at what the job required. I’m really proud of all the work I contributed to there. But it’s also true that I missed the “gnarliness” of delivery – the practical, hands-on experience without which I wouldn’t have been as good at the job. So I always knew my time focusing on the conceptual and advisory wasn’t forever – it’s testament to the team and the content that I surprised myself and ended up staying for five years.

Continue reading

AI in government: it’s about people, not technology (as always)

It was our first week back for Vineyard English School after the summer break1. Many familiar faces were absent, but one young Eritrean was eager to see us – he’d just received a letter about his asylum claim.

We were back in the hotel today after stopping over the summer (more volunteers would allow for doing this year round). Here's a photo of a letter that had been received by one of the hotel residents. Two native English speakers had to check with one another that we actually understood it.

[image or embed]

— Benjamin Welby (@bm.wel.by) 11 September 2024 at 18:13

The letter was dense, bureaucratic, and impenetrable. It’s a far cry from the aspirations for content design that so many advocate for as a central plank in reimagining the relationship between the state and its users.

He looked to us for an explanation. But even among the fluent English speakers, we had to consult amongst ourselves to ensure we understood it correctly. Hardly surprising, since according to The First Word’s readability test, this letter is on par with reading Nietzsche.

A visual display of book covers arranged by difficulty level, ranging from "Very Easy" (0-20) to "Very Challenging" (61-100). The cover in the middle, labeled "20 - 30," stands out in yellow and features the title "Beyond Good and Evil" by Nietzsche. Other covers represent a range of genres and styles.

The power of AI

I reached for ChatGPT.

Continue reading

No Neeeeed to trust ChatGPT, especially for your nostalgia

A familiar catchphrase, an unexpected blank stare

Our family has just spent a lovely bank holiday weekend with friends enjoying the National Coal Mining Museum, a tremendous 3-1 victory for Bradford City, a delicious curry at The Sweet Centre, some good church at St Peter’s Bury, and the splendid surroundings of the National Trust’s Shugborough Estate.

As we were driving to our first rendezvous in Wakefield, our friends, who live near Bolton, sent a message apologising that they would arrive after us. I reacted on our end with a loud “No neeeeed”, a catchphrase remembered from my childhood, delivered in an attempted Lancastrian accent. Christine, ever quick to capture a moment, got me to repeat the impression and sent it back as a voice note.

They replied with their own note, which we couldn’t decipher. So when we met up we asked them what was going on. To our surprise, they had misunderstood our message – they thought we were saying “Welbeeeees” as a greeting and had responded with a cry of their own family name.

I was puzzled. I was sure “No neeeeed” was a common cultural touchpoint but instead it was met with blank stares. Confidently, I told them it was a catchphrase from Mark and Lard, the iconic radio duo of the 90s.

However, when I tried to back up my claim with a quick Google, I hit a brick wall.

Continue reading

I used ChatGPT to improve the speed and accessibility of my WordPress blog

Adding ALT text is a good habit to develop. I can’t say it has become second nature but I do try. And I have definitely found it to be a great use for ChatGPT.

Since ChatGPT got vision capability it’s a very simple thing to paste an image and ask for ALT text. It’s usually very good. Although I was shaken when it casually described me as middle aged in this image. I suppose it’s accurate now I’m 40 but it was brutal to see that text appear on the screen.

Yes, it is an important habit to develop and an important principle to prioritise. However, I think it should also be something where AI can solve a real life problem in a very practical way. It feels like Twitter, Facebook, WhatsApp et al should fold in this functionality and automatically add ALT text whenever you upload a new image.

Be the change you want to see

So with that thought in mind I was very aware that I recently published a series of blog posts consisting of 140 images of slides detailing how to build a data driven public sector, the majority of which did not include ALT text1.

Continue reading