# AI and Civil Service Reform This note collects reflections, use cases and ideas at the intersection of artificial intelligence (AI) and civil service reform. ## Entries [[2025-06-13]]: Discovered this [case study from Stanford’s Gordian Knot Center](https://www.linkedin.com/feed/update/urn:li:activity:7326316955152375808/) analyzes why two major U.S. Department of Defense initiatives - the Joint Artificial Intelligence Center (JAIC) and the Chief Digital and Artificial Intelligence Office (CDAO) - failed to deliver meaningful AI adoption. Both were created to modernize military operations through AI, but ended up bogged down by poor organizational design. JAIC lacked authority over operational units and budget enforcement; CDAO inherited fragmented teams through a merger that was nominal on paper but siloed in practice. Despite consistent Congressional funding, neither unit had control over how resources were used. JAIC was positioned under the Chief Information Officer, limiting its ability to mandate AI use across services. CDAO added governance structures like councils and compliance frameworks, but these created decision bottlenecks rather than enabling implementation. Innovation teams became "flyaway" advisors without tools or authority to act. Cultural resistance within military branches further undercut efforts, as services preferred to build their own AI programs rather than collaborate with outside units. **[[2025-06-03]]**: From September to December 2024, the [[UK Government Digital Service (GDS)]] led a large-scale trial of Microsoft 365 Copilot across 13 UK government departments, involving over 20,000 civil servants. The aim was to test whether AI tools could improve productivity and satisfaction. Results were positive: users reported saving an average of 26 minutes per day, with high engagement across most departments. 82% of participants said they would not want to return to working without Copilot, and satisfaction and recommendation scores were strong (7.7 and 8.2 out of 10, respectively). The experiment also highlighted a clear link between familiarity with AI tools and the level of benefit perceived. Users with accessibility needs, such as dyslexia and dyspraxia, reported that Copilot made tasks like writing and editing significantly easier, helping them improve tone and clarity. The availability of Copilot also had a positive effect on how people felt about their work: 63% said it improved their satisfaction with their employer. Mid-level staff (HEO to Grade 6) reported the most value, likely because they balance both operational and strategic tasks. On the other hand, professions that rely heavily on nuance and judgment, such as policy and HR, were more cautious in their use. **[[2025-05-24]]**: Came across the [AGI Social Contract](https://www.agisocialcontract.org/) project, and found it quite compelling. It's a growing initiative led by researchers from institutions like Google DeepMind, MIT, ANU, and the AI Objectives Institute, aiming to rethink the foundations of economic governance in light of transformative AI systems. What stood out to me is their focus on the role of governments in managing the economic shocks of AGI - from labor displacement and taxation, to public ownership and global coordination. They're not just raising abstract questions, but publishing concrete proposals and organizing a summit in Stockholm (August 2025) to bring together policymakers and scholars around these issues. **[[2025-05-21]]**: I’ve compiled a list of the main points from the report that stood out to me. The white paper _[The Agentic State](https://media.licdn.com/dms/document/media/v2/D4D1FAQG_6W3L6DTDxg/feedshare-document-pdf-analyzed/B4DZcQZbgiIEAc-/0/1748326801617?e=1750291200&v=beta&t=sL6W7yZ5dcpgNrZpnNj7wRRy0Rz87ImRMSP-CCFrBmA#page=26.08)_, published in May 2025 by the Berlin Global Government Technology Center, points out that successfully adopting artificial intelligence in government depends not only on technology... but also on people! Though not framed as a traditional HR or civil service reform document, the report's core message is clear: AI-driven transformation will fail without significant changes in how governments recruit, develop, and lead their workforce. The report highlights the inadequacy of current hiring systems, which are built around rigid credentials and lifetime careers. This makes it difficult to attract and retain top AI talent. The report also points to cultural and structural barriers, such as limited career paths for technical specialists and a risk-averse, process-driven environment, that actively hinder innovation. In short, the report suggests that building an "agentic state" is as much a human capital challenge as a technological one. **[[2025-05-07]]**: The [[Tony Blair Institute for Global Change (TBI)]] is the one publishing some of the best insights on the application of AI in government. Period. In May 2024, they released _[Governing in the Age of AI: A New Model to Transform the State](https://institute.global/insights/politics-and-governance/governing-in-the-age-of-ai-a-new-model-to-transform-the-state)_, with [[Tom Westgarth]] among the authors. In that report, they propose giving every citizen a digital public assistant for tasks like tax returns, embedding AI agents in civil service teams to speed up work and red-team documents, and using digital policy twins to simulate decisions safely before they go live. Today (07 May 2025) they published _[Governing in the Age of AI: Reimagining Local Government](https://institute.global/insights/politics-and-governance/governing-in-the-age-of-ai-reimagining-local-government?utm_source=social&utm_medium=tbi_staff&utm_campaign=governing_ai_local_gov)_, with [[Alexander Iosad]] on the author list. This follow-up applies the same AI tools at the subnational level: introducing AI co-workers to cut backlogs, piloting local navigation assistants to help residents navigate services, and building data-driven planning platforms to free up staff for more complex work. **[[2025-05-04]]**: Govern for America, along with [PIT-UN](https://pit-un.org/) and the[ Center for Public Sector AI](https://www.cpsai.org/), published a [report](https://www.govforamerica.org/ai-tech-talent) showing that while many U.S. states are exploring AI, most efforts are still experimental and rely on volunteers, with key challenges including hiring skilled talent and measuring real impact. **[[2025-04-11]]**: Ed de Minckwitz has been appointed as Senior Policy Fellow at [[Policy Exchange]], where he will lead a new programe titled [Government in the Age of Superintelligence](https://www.cityam.com/can-ai-do-it-better-why-starmer-is-right-to-reimagine-the-civil-service/). The project will explore how AI could reshape the civil service. **20/03/2025**: The [UK’s Open Innovation Team](https://www.gov.uk/government/organisations/open-innovation-team) is designing a new AI adoption programme for the Civil Service, with a focus on deployment rather than just tool development. [Ben Taylor](https://www.linkedin.com/feed/update/urn:li:activity:7302739589830508544/) (Deputy Head) argues that most current efforts overlook the “techno-social” aspects of AI - like training, workflow adaptation, and behavioral change. Drawing on historical insights, he highlights the messy, slow, and non-linear nature of tech adoption in government. The team is planning a pilot with the Department for Education to explore use cases, tool fit, legal questions, and productivity metrics. **05/03/2025**: In September 2023, Pennsylvania became [the first U.S. state to partner with OpenAI](https://www.cityandstatepa.com/politics/2024/01/pa-becomes-first-state-partner-openai-ai-pilot-program/393218/) through a generative AI pilot using ChatGPT Enterprise. The program ran until March 2025 and involved 175 state employees across 14 agencies. [Preliminary results](https://www.pa.gov/content/dam/copapwp-pagov/en/oa/documents/programs/information-technology/documents/openai-pilot-report-2025.pdf) show that 85% of participants had a positive experience, estimating time savings of up to 95 minutes per day. ChatGPT proved most useful in public-sector tasks like drafting job descriptions, summarizing policies, and brainstorming communication strategies. Use cases clustered around three roles: _innovation engines_, _bureaucracy hackers_, and _strategic communicators_. ^pennsylvania-openia-partnership **01/03/2025**: [InnovateUS](https://innovate-us.org/) is a nonprofit initiative in the United States that provides free online training for public sector professionals in areas such as digital technologies and artificial intelligence. The program is developed and managed by [The GovLab](https://thegovlab.org/), a research center based at New York University (NYU). InnovateUS offers self-paced courses, live online workshops, and coaching programs aimed at professionals working in local, state, and federal government agencies. The initiative is funded by organizations such as [[Google.org]], [[Rockefeller Foundation]] and [[Schmidt Futures]].