AI and public services: our approach
“The test of AI in public services is not how smart it is but what it adds and how well it serves us.”
At dxw, we always start with the service, not the technology. We ask why first, then how AI could help. We’ve been doing this in real public service settings for years, since well before generative AI arrived.
Below, we set out why AI is so alluring, why it can’t rescue public services on its own, and what that means for how we use it.
Reducing the friction in public services
In the effort to ‘build a better Britain’ and boost national productivity, it’s easy to be dazzled by the lights of AI.
Public services exist so that citizens can go about their daily lives with the least friction.
People and organisations want to get stuff done, now and in the future. Public services are there so that can happen. Now? — to get a passport, say, or check school Ofsted reports, or register a new business, report a pothole. And in the future? — we all benefit from a safe, healthy, well-educated population, with good laws and governance (and roads without potholes!).
There’s always an urgency to reduce the friction and make it easier for people and organisations to get on with things. But for a while now we’ve had to face strong headwinds that have come at us from multiple angles. So, as a nation, we’re looking for all the ways we can to get a boost.
AI is alluring
The messaging of the AI labs is that artificial intelligence is the key to unlocking productivity. In 2023, McKinsey predicted AI had “the potential to automate work activities that absorb 60 to 70 percent of employees’ time today.” In 2024, an IMF report calculated that “almost 40 percent of global employment is exposed to AI”.
This is all very appealing, seductive even. Companies like IBM and Oracle and Cisco and Cloudflare and Dropbox and Block and many others have made dramatic job cuts through their AI adoption. The same (or greater) income with fewer people is immensely attractive. The appeal for the state is similar, especially when national productivity has been so frustrated by those headwinds. The launch of the Artificial Intelligence Playbook for UK government in early 2025 was heralded with an extremely bold position:
AI is at the heart of the UK Government’s strategy to drive economic growth and enhance public service delivery — it offers significant opportunities to reduce costs and enhance public service delivery across government.
Launching the Artificial Intelligence Playbook for the UK Government, 10 February 2025
But AI can’t rescue us from ourselves.
In February this year, Gartner published a report predicting most companies making AI job cuts will rehire within the year. AI will not rescue productivity from a trough. It can handle volume. It can do repetitive tasks determined by clear rules. And it will create stuff almost inexhaustibly. But it’s very different from humans.
In the public sector context, this is every bit as true.
AI cannot rescue public services from themselves.
(With appreciation to Tom Loosemore at Public Digital for initially articulating this in their context.)
AI is an amplifier
The I in AI is deceptive.
AIs are elaborate prediction machines. They use statistical probability and rapid iterative, pattern-matching to quickly generate coherent text, code, or images based on vast amounts of data. But AIs lack a world model or any real understanding of it.
AIs don’t know what a handlebar on a bike actually is, as a thing in the world, what it feels like to hold one, or the difference between holding the handlebars of a balance bike, a shopping bike, a mountain bike, a racing bike, a clown bike, a pennyfarthing, a unicycle, or a motorbike.
A bike designed by AI.
Statistical probability combined with its lack of a world model means that AI amplifies things.
If you’re already working efficiently, with a tight focus, strong empowerment, with good source materials, close to users, learning fast… then AI will amplify all that. It might even help make things pretty amazing.
There are clear places where AI tools are game-changing – most obviously in identifying diseases or in financial risk management. And more prosaically, in giving access into the vast amounts of unstructured data swimming about the place.
But, on the other hand, if you have a spaghetti mess of governance, isolated silos with rigid handoffs, top-down assumptions, all under centralised control… then AI will just make the tangles even messier.
More succinctly: in public services, as in all human endeavour, it matters where things fail.
AI is great when you’re focused on how things work; it’s not great at understanding how things go wrong. If you only care about how things work, AI is amazing. But if you need a bicycle with a brake that is a brake not a pannier rack, you need to understand the real world.
AI is additive
At dxw, we view AI as a thing to be added, rather than applied as a substitute.
Our expertise is in making digital solutions that work for public services. Or said differently, for us the point of the digital solutions we make is to enable public services. We identify and then execute the right things to get the stuff done that makes public services work. And now AI is a part of that.
When AI is considered for addressing a problem, our focus is on the service that’s being delivered. The questions we ask are about why, first, then how AI could be added to the service being used by the public and the team that’s running it. Working efficiently, with a tight focus, strong empowerment, with good source materials, close to users, learning fast – to make it better, or help it to do things it’s not been able to do before.
Our focus is squarely on how AI can help public service users and support the teams delivering them, not using it as a substitute for people or other things as a shortcut to efficiency.
AI should be added onto good practice, not used as a fudge for good practice. The test of AI in public services is not how smart it is but what it adds and how well it serves us.
AI, public services, and dxw
We’ve dedicated a lot of effort to AI experimentation, research, delivery, hiring, etc., in the same way we have with all the new technologies we’ve encountered since we were founded in 2008.
Like AI, each one has offered a novel way to do things. Many have had a bubble of hype around them. Many of them have had risks. Some, like cloud computing and containerisation, have totally transformed how we work. Others haven’t. AI is a new technology, but its newness isn’t new.
We’ve accumulated deep experience in applying AI in real-world contexts in public services in central government and beyond. Stemming all the way back to AI 1.0, before the current era of generative AI.
We understand how to get the best out of AI and how to avoid or ameliorate the worst; whether in identifying, developing, delivering and running them, or AI components embedded in the service itself, or the transformation (and its challenges) needed to get there.
We match AI with the purpose of public services.
——————————————————————
Appendix: background reading
Goldman Sachs AI “could drive a 7% (or almost $7 trillion) increase in global GDP and lift productivity growth by 1.5 percentage points over a 10-year period.”
Dario Amodei, CEO of Anthropic “Cancer is cured, the economy grows at 10% a year, the budget is balanced — and 20% of people don’t have jobs.”
Jim Farley, CEO of Ford “Artificial intelligence is going to replace literally half of all white-collar workers in the U.S.”
OpenAI’s CEO, Sam Altman “certain job categories” will disappear.
Meta makes 2026 the year to turn their workforce “AI native,” aiming to slash team size across the company with targets of up to 60% reduction.
Amazon announces 16,000 corporate job cuts in its push into AI adoption.
Klarna lays off staff, then hits reverse.
Commonwealth Bank of Australia does the same thing.