10 lessons from 2025

I began the year working at the Incubator for AI in Cabinet Office and then GDS, and finished it at Digital Prevention Services at NHS England.

10 things I learned:

1. The centre is difficult to be in and to be outside
2. The French are way ahead of us
3. Your job is a costume you may choose to take off
4. You’re lucky if you find your people
5. Hard skills teach good taste
6. NHS England doesn’t know how to understand itself
7. Work is just a lot of variously-formalised social relationships
8. “Scale” is an idea worth questioning
9. I know nothing about medicine
10. What works is rehearsing the context

The centre is difficult to be in and to be outside

The Incubator has a unique remit to roam across government and focus the attention of the centre (ie No10 etc) on teams and technology in departments. This meant we could work on genuinely cross-cutting stuff like consultations. I passionately did not want to waste that opportunity, and AI was interesting.

As time wore on I often found myself at the more sceptical end of AI opinion in the team. That tension was productive, but tiring. And seeing how things would stall or crumble once the focus moved on, I came to be critical of the centre itself, which can be whimsical, impatient and excessively focussed on technology.

In my new job we are eager for the centre to help, but not like that. As Sarah Fisher notes, teams delivering inside NHS England still encounter difficulties that simply don’t exist elsewhere in government. We would welcome support from a more engaged GDS.

The French are way ahead of us

In June I went to Paris to learn about (and hack a bit on) “Le Suite”, the French Government’s set of open source software to replace Microsoft Office etc. This was timely: not long after the Microsoft/Hague controversy and AI sovereignty was a hot topic too. I noticed:

Firstly, while France, Germany and the Netherlands were all well-represented at the hackathon and have their own offerings in this space, the UK is nowhere in this conversation. GDS is still legendary, but more for past glories; it’s not interested in this rather bracing new ambition of disentangling entire government estates from big tech.

Secondly, there are some embarrassingly basic chunks of Digital Public Infrastructure the French have already got in place but we’re nowhere near, like single sign-on for public servants. (This was a particular source of pain and complexity for i.AI as we tried to ship pilots to different departments and had to onboard every single one separately). Good luck to the GDS team now working on this.

Thirdly, it was instructive to see a lot of commercial pitches at the hackathon from companies selling hosting platforms and other infrastructure back to teams who adopted Le Suite. In a way this was an encouraging example of the state more forcefully shaping the technology market. On the other hand, every proprietary feature these companies introduced to compete with one another would further fragment the consistent, open ecosystem the tools were intended to create.

Finally, by centralising demand early, the French have a path to safer, greener, opener, cheaper AI for the state as they need it. At the incubator I regret not getting to work on what I still see as a potentially transformative project: centrally-assured, centrally-procured, AI-as-a-service on the model of GOV.UK Notify. At the time I left, a team was going to work on this. I hope that happens.

Your job is a costume you may choose to take off

In leaving this job (and jobs before it) I experienced a tension that’s roughly “I’m giving up early, I’m a quitter” vs “I’m becoming an expert in gazing into the abyss, and that’s a life I don’t want”. I observed this time around that both these objections are concerned with who I am, not necessarily what I choose to be responsible for at this moment. You are not personally responsible for the fate of the world, and you do not have to ritually punish yourself in order to learn and contribute. (You may even find that laying off the ritual punishment helps you to learn and contribute more!)

You’re lucky if you find your people

It’s a trope that quests end by discovering something of yourself, not of the world. So with me, venturing out to i.AI for just over a year and a half and then rejoining former DfE colleagues and others at NHS England.

It’s hard to describe the weirdly intense pleasure of seeing two people you know and like working together for the first time. There is belonging in that, which seems to me to contain more power and potential than any structure a disinterested organisation can construct.

This has made me happy. But it’s also powerful: these strong work/social ties are the way we’ll build the necessary team coherence for coming to grips with the huge challenges our work and NHSE itself present.

Most of all I learned by contrast that this extraordinary, unlikely state of affairs did not occur by accident. Rachel Hope created it.

Hard skills teach good taste

For the past 5 years, at DfE and then i.AI, I have become more and more detached from the code.

When I joined NHSE in August my job went from being a leader in a team of ~70 to a team of ~600, and I finally let go of the last of it. The plane has gone up into the sky. It’s weird up here, and quiet.

It’s weird because I’m here because of my skill in an activity I no longer practise professionally. So I now fantasise about going amongst the people, taking a job as a tech lead for a few months. I often think I probably ought to do that, and probably soon, before I become even more detached from reality. And yet I must have scope. And yet I must have depth. And yet I must do all this other stuff too. And yet and yet and yet.

I have to believe that there is something I have learned that enables me to understand and support the right things, even if I am not actually doing the doing. I think maybe it’s taste?

NHS England doesn’t know how to understand itself

And there is different work here, too.

There are so many things—systems, structures, relationships, histories—I don’t understand. I don’t really like any of the tools we have to make sense of them. Software architecture, enterprise architecture, ideas like moldable development which briefly caught my eye earlier this year.

Most of all, though, I am deeply suspicious of the language of “value” and “benefits” that structures this huge organisation. To watch the construction of business cases is to watch so many divisions contorting themselves into legibility for senior leaders and for Treasury, trying to find expressions that unlock investment by offering up a return.

It is a brutal coincidence that the language we use to ask for money is the same language we must use to define our purpose and make sense of our work. But politics demands it, just as it demands the constant accounting which puts a hard ceiling on trust up and down the hierarchy. The system is against us.

But we don’t have to believe in it. My instinct is to drive in the opposite direction, away from centralisation and from command-and-control, and towards independence, federation and autonomy.

Work is just a lot of variously-formalised social relationships

What seems to work for me now is talking to people—a lot.

Two articles in particular from this year have helped me develop that practice and given me more confidence in it.

One was this podcast by Bruce Daisley, where Alex Haslam argues that social identity is the overwhelming factor determining the success and efficacy of teams. I think almost daily of this observation:

The leader of the London Fire Brigade is a working class man who’s come through the ranks. And so when he stands up there and he speaks in the accent that all of the colleagues speak and he looks like one of them who’s has ascended and he behaves in that way.

The other was this metaphor from Steve Messer:

Many hands, many tillers.

So I now see a big part of my job as this, from the Daisley/Haslam podcast:

[that’s what] leadership is about. Social identity management. Fundamentally, that’s what leaders are. They’re people who bring us into groups, who keep the groups together and direct the group forward.

The groups can (and must, at 600 people) be heterogenous. But I am not worried that it is impossible to create a cohesive culture in large groups. The fact that the NHS exists at all proves that it is possible.

“Scale” is an idea worth questioning

It’s a funny paradox that technology promises us as leaders a panopticon for us to observe and control all activity in our organisation, whilst simultaneously offering everyone the means of creating practically unlimited complexity in all directions.

These compressions: business cases, enterprise architecture—future vestiges, I think, of what Hilary Cottam calls the “industrialised” model of public services—are technocratic fantasies of which “scale” (per the panopticon) is the aim, and of which scale (per the exploding complexity) makes a mockery.

As Anna Lowenhaupt Tsing puts it:

Scalability requires that project elements be oblivious to the indeterminacies of encounter; that’s how they allow smooth expansion. Thus, too, scalability banishes meaningful diversity, that is, diversity that might change things.

In 2026 I look forward to arguing in favour of the “indeterminacies of encounter” via rigorous approaches to culture, ways of working, and open standards in the work we do, and in the way we work.

I know nothing about medicine

One striking observation I heard this year was at the October Breast screening away day, where an NHS England director with a clinical background said “risk stratification is the bread and butter of medical practice”.

A few weeks later I heard another striking observation from Jessica Rose Morley, accompanied by a slide referencing numerous scientific studies: “when we went and did a systematic review of what happens after [risk stratification tools] have been deployed, it transpires they don’t work”.

Given our job is prevention, it would be good for us to be able to talk about that interesting conflict, and perhaps to be able to adjust our work in light of it. But it’s difficult to do that when medical knowledge is concentrated in the heads of “clinical” colleagues. Indeed, the relationship between clinical and digital teams in NHSE is conventionally one of governance, rather than of direct collaboration—a matching pair for the buy-then-govern model of big IT procurements.

If we’re ignorant as well as influential, it’s our duty to mitigate that. And whilst it would be extremely fun and interesting for us all to do medical degrees, there isn’t time. So there are two DPSP projects I’m sponsoring in 2026 I’m particularly excited about.

One is collaborating with academics on jointly making sense of our work, building a strong evidence base for “test and learn” approaches as well as finding the best ways to measure, understand and articulate the benefits of specific programmes.

Another is developing the discipline we’re tentatively calling “DevClinOps” (snappier name suggestions welcome)—an evidence-led approach to shortening the feedback loops between clinical assurance and product teams, with a long term view to converging “clinical” and “product”.

What works is rehearsing the context

This was the year I gave up on computerised note taking. I write all my notes longhand. I experimented with keeping manual indexes but that was busy and not that useful. What is useful is continually rehearsing the context. Search and graphs and all the rest of it offer mainly prevarication.

Other stuff

A few ideas I found useful this year.

Assume everyone radically underestimates everything all the time. It’s usually true, and I find it just about balances out my own Pollyanna-ish enthusiasm to Start Things.

Most prioritisation problems can be reframed as sequencing problems. Nobody can see into the future, so a good way of declining to do something is to tell them you’ll do it next. You may find that when the time comes, you do want to do it—or that they don’t.

I came across Robyn Dawes’s observation that “cognitive capacity shuts down in the absence of a story”. Mine certainly does. Now I ask of everything: what’s the story so far, what’s the end, what happened last, what happens next. The path to success is going to be through telling more and better stories.

Surprisingly often, people don’t aspire to do, or indeed don’t have, the job role they’re actually best at.

In an organisation where everyone wants to do good, people can associate success with personal virtue in a way that does not necessarily foster collaboration.

Things and systems that are worn, shaped to you, that fit you and you fit: these are deep repositories of (for want of a better word) meaning. It’s true they can collapse, fall into comfort, or go out of date. Even so, they can also contain an order of magnitude more potential than starting over.