PD Newsletter #104: Don't risk automation without redesign 🤖

👋🏽👋🏻👋🏾 Hello, welcome.

In this edition: the four levels where change happens in organisations, how governments can build state capacity, and why AI generates more demand than organisations - public or private - can handle. Plus, we're hiring!

Copies of our free book, offering a guide to building technological adaptability and resilience, are still available. Order yours now.

Rosemary

🌟 View this newsletter on LinkedIn

🚀 Forwarded this email? Sign-up.

Latest from Public Digital

🗓️ We're sponsoring TransformGov Talks. Their next event is tonight at 6pm GMT, in person or online, featuring Liam Hawkes and Violette Gadenne, (Policy and Innovation Lab, Home Office) and ​James Plunkett (Kinship Works).

Read our reflections from:

🌱 UKGovCamp 2026, where we write about some of our favourite sessions from the day

🇨🇦 PD Day FWD50, where Mike Bracken and David Eaves spoke about how governments can build digital sovereignty.

🎨 Finally, our CEO Ben Terrett CBE has been announced as the Chair of Art UK.

Internet era ways of working

🏛️ A long read on the work of creating a shared intranet at the V&A, from V&A Redesign lead, Lizzie Hines. The project involves creating a single, searchable source of truth for staff as the organisation shifts to supporting a larger workforce, working across multiple sites. Brilliant stuff on starting with principles, not pages, and approaching tech choices with the question ‘is it good enough?’ rather than ‘is it perfect?’.

📶 Alistair Ruff and Katherine Wastell examine the four levels at which change happens in organisations - and explain why being able to articulate them is so important for leaders: “A strong understanding of these levels is what distinguishes mature digital leaders from those trying to implement off-the-shelf frameworks”. Plus, the MacGuffin Pattern from Tom Eldridge provides a framework for service designers who want to change systems, not just optimise within them.

🪄 John Cutler on the impact of AI on leadership assumptions about context and control. “If vendor narratives are taken at face value, context becomes a kind of universal solvent: pour in enough of it and magically you get clarity.”

🥗 Fantastic stuff in AI recipes for charities, offering “practical approaches to using AI and machine learning in your charity.” By design and innovation agency, Make Sense of It.

🏷️ A helpful AI taxonomy, from Narain Jashanmal. Don’t say: "We're using AI for this." Do say: "We're using Analytical AI to score, then Generative AI to personalize."

Technology in focus

🍾 Responding to Matt Shumer’s viral essay on AI and its impact on jobs, David Oks counters the doomerism with the view that human labour is safe (for now) thanks to ‘human bottlenecks’. “No matter the level of AI capabilities, we should expect a real and powerful complementarity between human labor and AI, simply because the “human plus AI” combination will be more productive than AI alone.” That argument may also explain why AI has reportedly contributed little to economic growth.

🗣️ Google has partnered with several African research institutions to launch WAXAL, an open-speech dataset for 27 African languages. At the same time, Microsoft has released its own voice recognition tool for 39 languages. Both projects reflect the reality that voice, rather than text, is how many people in Africa will encounter AI.

🦞 Loads of experimentation happening with OpenClaw, from using it to automate sauna bookings (reducing “small repeated frictions… exactly what automation should target”), to building a benefits calculator chatbot. Tom Loosemore points out it’ll be some time before the security of these agents is good enough for them to be mainstream - time that governments must spend working out how the public sector will respond. As Martha Lane Fox writes: capability is moving faster than comprehension.

🎭 Strange reading in the best of Moltbook - the Reddit-clone social network for OpenClaw agents (or, as it turns out, some humans) to talk to each other about subjects including the nature of consciousness, and keeping glitches as pets. As one MIT Tech Review article argues, it’s peak AI theatre: “less like a window onto the future and more like a mirror held up to our own obsessions with AI today.”

💊 The largest user study of language models for medical self-diagnosis so far has revealed that you’re better off using Google search and your own judgement - rather than chatbots - to self-diagnose. The inaccuracy of Google’s AI overview is made worse by its downplaying of disclaimers and the way its design means it is perceived as a single source of truth.

🇪🇺 What happens when you try and build a start-up exclusively on European infrastructure? “It’s effort worth spending. But it is effort.”

Digital government news

📈 “The risk is not automation per se, but automation without redesign.” How AI generates demand rather than creates efficiency for governments - reducing friction and exposing higher numbers of citizen demands, without adapting institutions to meet them. A similar trend is appearing in the private sector. It speaks to the tendency for AI adoption to make a bad service efficient, rather than making a service effective. This report from Ann Lewis at the Niskanen Centre argues similarly: government AI adoption will only work when accompanied by effective operating models.

🇪🇪 It will be interesting to see how Estonia’s new Digital Agenda tackles that problem. It commits to transforming services through AI akin to the way they were transformed through digital. “The goal is not more AI projects. The goal is that every service is simply built with intelligence by default.”

👑 Quebec's ambitious new digital sovereignty plan is a welcome step for the province, but will prove challenging given its continuing reliance on US contracts - several of which are plagued by delays and cost over-runs.

🏗️ A long read from Andrew Greenway on how the UK government can build state capacity, and why it badly needs to - building on his writing with Jen Pahlka. “A government that is serious about state capacity must do 2 things, urgently and in parallel: one: fundamentally reboot the character of the ‘permanent’ civil service…. and two: create new institutions.”

✨ Learnings from two years building with generative AI in government, from Alfie Dennen in the UK government’s Department for Business and Trade.

💬 How user support helped Notify send 12 billion messages since 2016. “User support is more than just answering a question or helping users complete a task – it’s a shared responsibility, a source of insight, and a catalyst for improvement when delivering a service.”

Some fun things

🐦 An uplifting blend of ornithology and data in Searching for Birds.

🚶‍♂️ Walk the internet archives with this random website generator.