The service comes first: AI in government
At dxw, we see AI as something to add to public services, not a replacement for people or good practice. Used well, it makes services better for the people who use them and the teams who run them.
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.
Start with the service
Public services exist so people can get on with their lives with as little friction as possible. Getting a passport. Registering a business. Reporting a pothole.
Making that easier matters, and the UK government's AI Playbook puts AI at the heart of its plans to improve public services and reduce costs. We share that ambition. But the best results come from starting with the problem people face, then working out what will genuinely help. Sometimes that's AI. Sometimes it isn't.
AI amplifies what's already there
AI is very good at some things. It handles volume, works through repetitive tasks and makes sense of vast amounts of unstructured information.
But AI is a prediction machine. It matches patterns without understanding the world. Ask it to design a bike and you might get something convincing, until you notice the brake is a pannier rack.
That's why we think of AI as an amplifier. Add it to a team that's focused, empowered, close to its users and learning fast, and it makes all of that better. Add it to tangled governance and rigid silos, and it amplifies those too.
As Tom Loosemore at Public Digital put it, AI can't rescue public services from themselves. But it can help good services go further.
Add AI to good practice
So we add AI to services. We don't use it as a substitute for people, or as a shortcut to efficiency.
When a team is weighing up AI, we look first at the service: who uses it, who runs it, and what they need. Then we look at where AI could make it better, or help it do things it couldn't do before.
We pair that with the practices that make any digital project succeed: a clear focus, empowered teams, good data, closeness to users and fast learning. AI works best on top of those foundations.
For us, the test of AI in public services isn't how clever it is. It's what it adds, and how well it serves people.
New technology, familiar territory
We've worked with plenty of new technologies over the years, many arriving on a wave of hype. Some, like cloud computing, changed how we work for good. Others didn't.
AI is new. But its newness isn't. We bring the same test-and-learn approach to it.
At the Department for Education, we ran a 3-month experiment as part of a joint team to see whether an AI assistant could help expert officers answer colleagues' questions faster. It couldn't replace their judgement on local context. But it was a strong researcher, finding the exact guidance they needed. The human stayed the decision maker.
For the Office of the Patient Safety Commissioner, we used AI to rebuild their website in 2 weeks, moving it off a restrictive page builder. One agent made changes, another checked them against the live site, and our team judged the quality. We removed 56,000 lines of legacy code, and the PSC team now manage their own content.
For us, that's what AI is for: making good public services better.
