# Rewiring the State Eoin Mulgrew, No. 10 (Downing Street) | AI Engineer Europe 2026 | 28:18 Source: https://www.youtube.com/watch?v=ObNKGf9YR0g Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks. Page: https://aietalks.com/talks/rewiring-the-state Published: 2026-05-18 Tags: enterprise, healthcare, team-adoption ## TL;DR - 10 DS is building a small, politically backed unit that recruits technical people from outside the civil service and gives them a mandate to ship inside government. - Forward-deployed engineers have built tools for policy simulation, legal analysis, delivery red-teaming, public reporting, and frontline services within weeks. - The fellowship is intended to create proof points that can change how the wider civil service works, although Eoin Mulgrew says the current model is still a workaround rather than business as usual. ## Summary Eoin Mulgrew describes how the 10 Downing Street data science team is building an AI engineering capability inside government. Its fellowship recruits exclusively from outside the civil service, pays close to market rates, selects for technical ability, and gives engineers political backing to work directly with departments. Engineers embed with lawyers, policy teams, communications staff, and operational teams, then take ideas into use within weeks. Examples include a policy simulation tool, a statute-book analysis tool that replaced planned external legal work costing £1.5 million, and a delivery red-teaming tool for Number 10. The team has also helped create public dashboards, a rapid public service, and projects across AI safety, planning, education, prisons, and criminal justice. Mulgrew is clear that a small unit cannot reform the whole state by itself. He sees the program as a pilot that must produce proof points before wider government processes change. ## Key ideas ### Government has large service problems that create room for technical intervention [01:53](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=113s) Mulgrew frames government as a 400,000-person industry rather than a single organization. He points to 7.25 million people on NHS waiting lists, around 350,000 court cases in backlog, and only one in five planning applications being decided on time. He cites a Tony Blair Institute estimate of a £40 billion annual productivity opportunity from AI in government. The barriers include low pay, hierarchy, bureaucracy, slow decision-making, and the safeguards required when government is accountable to the public and Parliament. These conditions make it difficult to attract people who want to work quickly on large problems. ### 10 DS uses an insurgent model with political backing and freedom to recruit [05:25](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=325s) The team responded by setting up a small unit at the center of government with a mandate from Number 10. It has unusually strong political backing to enter departments and get work done, a high degree of autonomy, and the ability to pay market rates within reason. Mulgrew says people may accept a pay cut if the work is interesting and the environment lets them do their best work. The recruitment process is separate from standard civil service hiring and is designed around technical skills. Its success rate is about 0.7% to 0.8%, and the program recruits exclusively from outside the civil service. ### The fellowship brings outside technical talent into government as missionaries [06:50](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=410s) The team has recruited people from AI labs, big technology companies, research institutes, Y Combinator companies, and serial entrepreneurship. Mulgrew says the program wants missionaries rather than mercenaries. Salary matters because joining government has to be economically possible, but money alone will not keep someone working when the job becomes difficult. His argument is that decisions reaching a minister can affect important parts of the country, so people with strong technical backgrounds may want to join if they are given a direct path to useful work. ### Forward-deployed engineers work inside teams and ship in weeks [09:05](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=545s) 10 DS handles simple use cases itself, especially work that can save money or improve delivery within a few days. Its forward-deployed engineers instead embed with policy advisers, lawyers, communications teams, pollsters, and operational staff. They watch how people work, find pain points, co-design solutions, and put new capabilities in users' hands quickly. More complicated backlogs require a partnership model, where engineers spend longer periods inside another department. Mulgrew says the first forward-deployed engineers in Number 10 are already working this way. ### Small tools can replace slow and expensive external work [10:18](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=618s) A policy simulation tool lets teams test the likely effects of decisions before they are made. One example examines how different Universal Credit choices could affect household finances. Mulgrew says it does not replace human analysis, but it lets more decisions use modeling at a faster rate. Another project addressed an intended £1.5 million purchase from an outside law firm to analyze the UK statute book. One engineer worked with in-house lawyers for a couple of weeks and left them with a tool they can run whenever needed. The result saves money and avoids analysis becoming outdated as new laws and regulations arrive. ### Delivery red-teaming and public dashboards add scrutiny inside Number 10 [12:22](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=742s) The team built a delivery red-teaming tool that is used every day by Number 10 delivery teams. It acts like a project management office in their pockets. It can inspect delivery reports and provide a second judgment about the departments producing them. The tool can flag patterns such as optimism bias, repeated amber risk ratings, and mitigations that usually fail. Mulgrew connects this work with government transparency. Until recently, the government had not published a public-facing delivery dashboard. It has since published two, including one tracking progress on the AI Opportunities Action Plan. ### The program is spreading through new teams and public-sector partnerships [14:45](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=885s) 10 DS has supported the AI Safety Institute by placing fellows there, including Harry Coppock, who worked on its Inspect tool for testing autonomous AI agents with tools. Former fellows also helped form the Incubator for AI in the Department for Science and Technology. The Incubator's Extract tool, built with DeepMind on Gemini, digitizes parts of planning applications, including handwritten material and hand-drawn maps, and is being rolled out to local authorities in England. The team is also helping evaluate AI tutors, with safeguards and benchmarks that include the cognitive load placed on students. ### The pilot must eventually change normal government operations [20:10](https://www.youtube.com/watch?v=ObNKGf9YR0g&t=1210s) Mulgrew says the unit is an experiment and admits that a small team cannot turn the whole civil service around on its own. The arrangement with ministers was to remove constraints, build a central proof point, and show what is possible. He wants the approach to become normal business across government. Over the next 12 to 24 months, he would like the team to target horizontal processes that affect large groups of workers, such as transcription for police and call-center operations in the Department for Work and Pensions and HM Revenue and Customs. The current model is a hack around the system, so wider change is still required. ## Notable quotes - "We're setting up a new team. It operates with a mandate from Number 10." (05:50) - "We want to recruit missionaries, not mercenaries." (08:26) - "Now we've got this tool that team can use and they can do it whenever they want at the drop of a hat." (12:03) - "This at the moment is basically a hack to get around the system." (23:54) - "Come join us, and we'll give you the keys to the state and see what you can do." (19:59) ## Tools & references mentioned - 10 Downing Street data science team - 10 DS - Tony Blair Institute - Calum Beer - AI Safety Institute - Harry Coppock - Inspect - Incubator for AI - Just AI - Dan James - DeepMind - Gemini - AI Opportunities Action Plan - Matt Clifford - Universal Credit - Y Combinator ## Who should watch - You work in government and need examples of engineers embedding with operational teams instead of waiting for large central transformation projects. - You are building a small public-sector technical unit and want a concrete model for recruiting, mandate, pay, and deployment. - You are an engineer or founder considering public service work and want to understand what the role can look like in practice. ## Related talks - [AI tools for Forward Deployed Engineering](https://aietalks.com/talks/ai-tools-for-forward-deployed-engineering) (Vasuman Moza & JD Pruitt, Varick Agents, 20:23) - [How Forward Deployed Engineering Is Done at Decagon](https://aietalks.com/talks/how-forward-deployed-engineering-is-done-at-decagon) (Sunny Rekhi, Decagon, 18:09) - [The Dirty Secret of Forward Deployed Engineering](https://aietalks.com/talks/the-dirty-secret-of-forward-deployed-engineering) (Natalie Meurer, Sierra, 16:49) - [What We Learned Deploying AI within Bloomberg's Engineering Organization](https://aietalks.com/talks/what-we-learned-deploying-ai-within-bloombergs-engineering-organization) (Lei Zhang, Bloomberg, 18:21) - [Dispatch from the Future: Building an AI-Native Company](https://aietalks.com/talks/dispatch-from-the-future-building-an-ai-native-company) (Dan Shipper, Every, 17:58)