Your Country Needs Builders

Government has a remarkable capacity to teach capable people that they cannot do things.

Some of the barriers are real. Public money and security deserve care. But spend enough time inside an institution and you also encounter constraints whose main authority is that somebody told somebody else they existed. Eventually, people stop checking.

There is a lot of learned helplessness in UK government, and it is learned for a reason. Spend months seeking approval for something you built in weeks, lose the person who agreed to own it, watch a reorganisation suspend hiring again: eventually, waiting starts to seem sensible. Capable people learn to expect very little of their own initiative.

I think we underestimate how much that can change when people are given the means to build something useful.

A bias to action

On my first morning as a No. 10 Innovation Fellow, I got my laptop at 9:30. At 10, I was in my first product meeting for Redbox, our AI tool for civil servants. I was an engineer being introduced, rather abruptly, to product management. There were mistakes to come.

That bias to action was why I loved the fellowship. I had colleagues who could help, real problems to work on and room to try. Those conditions matter more than another speech asking people to be innovative.

My degree apprenticeship let me put what I was learning to work straight away, with people I respected and problems I cared about. I have sought out that combination ever since. I want work that stretches me, and colleagues who help me see further.

That is part of what draws me to Dominic Cummings’s writing about bringing different skills into government. His call for “weirdos and misfits” included software developers, scientists and people with unconventional backgrounds, working alongside capable officials. There is talent that government struggles to recognise, and expertise already inside it that deserves better tools and more room to act.

Much of my fellowship went into helping build the Incubator for AI itself. As incubation lead, I worked on how we found worthwhile problems with departments and tested whether AI could help. That meant building a team capable of asking good questions, making something tangible and learning from the people who would use it.

It also meant being willing to discover that a promising idea was rubbish before spending another year on it. We were learning how to do this together. Engineers, policy officials, researchers and security specialists all had a part in making something work. The team was something we were building too.

What the institution keeps

Redbox gave us a useful lesson. We built it when access to frontier AI inside government was scarce. The models struggled with long documents and made substantial mistakes. Yet something happened around the software that I remain immensely proud of.

Our internal Redbox testing group numbered in the hundreds. The group chat was alive with people from HR, procurement and across the Cabinet Office arguing over model quirks, context lengths and prompting strategies with as much energy as any startup. They were trying things, comparing results and challenging one another. We had given them access to frontier models and a clear space to experiment. They supplied the curiosity.

The Civil Service can be remarkably innovative when it gets out of its own way. Those colleagues were becoming both inventive users and demanding customers, with enough experience to know when a tool was letting them down.

We eventually retired Redbox as other tools became available. Keeping pace with several enormous technology companies was never a sensible permanent ambition. The institution, though, retained something valuable: people who could judge what they were being offered from experience.

An in-house team that kills a project keeps the lessons. A product can reach the end of its useful life and still leave better judgement, stronger working relationships and people who know what to ask of the next supplier.

Mariana Mazzucato gives this an institutional explanation. Drawing on The Big Con, written with Rosie Collington, she argues that outsourcing can erode government’s ability to learn through practical experience. Building Redbox helped us become better buyers. We had learned enough to choose.

Colleagues and I have written a playbook about how the Incubator operated, led by Alexander De Ville. It records the methods and the lessons. What is harder to capture is how much I loved working with people who wanted to make the state work better, and were prepared to get on with it.

Do your bit

You don’t have to spend your whole career in government to leave it stronger. I have moved between government and industry because each has given me things to learn and bring to the other. There should be a way in, a worthwhile career for those who stay and a way back for those who leave.

A tour of duty can make a lasting contribution, provided you build with the people who will carry it forward. Shared ownership, code that colleagues can maintain and the confidence to tackle the next problem all matter. Much of what I hoped to leave behind at the incubator was a team that could keep discovering what was worth doing.

Government has obligations here too. If you are asking people to join, give them worthwhile problems, access to users, colleagues they can learn from and a credible route from a working prototype to a public service. Serving your country should be a sensible career choice. It cannot depend indefinitely on people pushing through obstacles in their spare time.

If that appeals to you, there are ways in. Explore the No. 10 Innovation Fellowship, including its open application route, and the Incubator for AI’s hiring page. Please don’t decide on their behalf that your background rules you out. Read what they are looking for.

We all have opinions about what the state should be able to do. I loved having a chance to make a small part of it more capable, alongside people who cared enough to try.

If you keep thinking someone should do something, I would love you to consider that it could be you.