AI is helping us build faster. But are we making better decisions?

AI is making it easier than ever to build things quickly. We can create prototypes, analyse information and explore solutions in a fraction of the time it might previously have taken. There’s some genuinely exciting potential in that, but I keep coming back to one question:

Are we getting any better at deciding what we should build in the first place?

It’s a question that stayed with me following our recent Evening with Methods roundtable, where we brought together people from across government to explore how AI might genuinely improve public services. What struck me wasn’t simply the enthusiasm for what’s possible, but the importance of being clear about the problems we’re trying to solve.

Having worked across different government departments and public sector organisations, I’ve seen how often teams face similar challenges. Different services and different users, but many of the same questions. We’ve accumulated years of research, established design patterns, technical standards and lessons learned, yet we still spend a surprising amount of time starting from scratch.

This is where I think AI could make a real difference. Not by replacing the expertise of researchers, designers or product teams, but by helping us find, connect and make better use of what we already know.

Imagine starting a discovery with a clearer understanding of what’s been tried elsewhere, what worked and where the gaps in evidence remain. We’d still need to question the quality and relevance of that information, but it would give us a better starting point and more time to focus on what we genuinely don’t understand.

The opportunity isn’t to replace the thinking. It’s to create more space for the thinking that matters.

I’ve seen the value of this approach in practice. In one public sector engagement, understanding sensitive user journeys, policy constraints and operational realities early helped create the conditions for more informed decisions before committing to a solution. In another transformation programme, reusable patterns and shared foundations meant teams could spend less time recreating the basics and more time focusing on the things that needed their expertise and judgement.

Neither was about moving quickly for the sake of it. Both reinforced the importance of creating space for meaningful learning.

Of course, being able to produce something quickly doesn’t necessarily mean we’re learning anything useful.

I’ve seen plenty of examples, long before AI came along, where an idea gains momentum and suddenly we’re discussing delivery plans before we’ve properly challenged the assumptions behind it. AI could make that even easier. We might generate a polished prototype in an afternoon, but if we haven’t understood the problem or identified what we’re trying to test, have we actually made progress?

For me, good product discovery has always been about reducing uncertainty. What do we believe to be true? What evidence would change our minds? And are we prepared to accept that our original idea might not be the right one?

A faster prototype isn’t necessarily faster learning. The value comes from what we discover, not what we produce.

Take a team exploring whether AI could improve an existing public service. They might discover that the real problem isn’t something AI needs to solve at all. Perhaps the guidance is confusing, information is fragmented or the process itself is unnecessarily complicated. The answer could be simpler content, a better process or joining up existing services. No chatbot or clever new technology required.

I’d argue that’s a successful discovery. We’ve avoided unnecessary investment and learned something that could genuinely improve the service. But there’s another challenge here, and it’s one I think we overlook. What happens when teams can learn faster than organisations can make decisions?

We can make it easier to gather evidence, but that doesn’t mean organisations are ready to act on it. I’ve worked in environments where teams have challenged assumptions, tested ideas and recommended a different direction, only to run into existing commitments, funding arrangements or governance processes that make changing course difficult. AI won’t fix that. It comes down to leadership, accountability and creating an environment where evidence can genuinely influence decisions.

It’s something I’ve written about before. Being product-led isn’t about delivering more; it’s about making better decisions about what matters. That means being willing to stick with an idea, change direction or stop altogether. None of this removes the need to understand users, consider accessibility, manage risk or apply the principles of the GOV.UK Service Standard. If anything, those things become more important as technology makes it easier to move quickly. Coming back to the conversations at Evening with Methods, I think there’s a genuine appetite to explore what AI could enable, particularly where it reduces repetitive work and gives teams more space to think, challenge and learn. But I’d rather measure its success by the decisions we’re making than the outputs we’re producing.

If AI makes it easier than ever to build things, surely we should be getting much better at deciding what’s worth building in the first place.