Rewiring the State — Eoin Mulgrew, 10 Downing Street

Key Concepts

• 10DS (10 Downing Street Data Science Team): A specialized, high-autonomy unit focused on integrating AI into government decision-making and public service delivery.
• Insurgency Model: An operational framework characterized by high political backing, rapid recruitment of external technical talent, and the ability to bypass traditional bureaucratic constraints.
• Forward-Deployed Engineers (FDEs): Technical staff embedded directly within policy, legal, and operational teams to co-design and implement AI solutions.
• Public Sector Productivity Crisis: The systemic challenge of high backlogs (e.g., NHS, court cases) and inefficient service delivery, estimated to have a £40 billion annual productivity prize through AI adoption.
• AI Safety Institute (AISI): The UK’s lead body for evaluating frontier AI models.
• Optimism Bias: A cognitive bias often found in government reporting where departments disproportionately rate risks as "amber" or underestimate project challenges.

---

1. The "Insurgency" Model for Government Transformation

Owain Mulgrew describes the 10DS team as a "scrappy little startup" operating at the center of government. To overcome the "oil tanker" problem of bureaucratic inertia, the team utilizes an insurgency model:

• Mandate: Operates with high-level political backing to bypass standard, slow-moving procurement and recruitment processes.
• Recruitment: Targets "missionaries" rather than "mercenaries." They exclusively recruit outsiders (from Big Tech, research labs, and startups) using a grueling, technical-focused selection process with a 0.7–0.8% success rate.
• Autonomy: The team is opportunistic, identifying high-impact areas where they can deliver results in weeks rather than years.

2. Methodologies and Operational Frameworks

The team employs two primary engagement models:

• Direct Implementation (Low-Hanging Fruit): For simple, high-impact tasks, the 10DS team builds and deploys solutions themselves. This includes workflow automation and internal tools.
• Partnership/Embedding Model: For complex, systemic issues (e.g., prison safety, planning backlogs), they embed FDEs into departments for prolonged periods to work alongside domain experts (lawyers, wardens, policy advisors).

3. Real-World Applications and Case Studies

• Statute Book Analysis: Instead of spending £1.5 million on external legal firms, an FDE embedded with in-house lawyers built an AI tool to analyze the UK statute book in weeks, creating a reusable, scalable capability.
• Delivery Red Teaming: A tool developed for 10 Downing Street to interrogate delivery reports, flagging "optimism bias" and assessing the effectiveness of risk mitigations in government projects.
• Project "Extract": A collaboration with DeepMind (using Gemini) to digitize handwritten planning applications and maps, aiming to improve the current 20% on-time decision rate for planning applications.
• Criminal Justice (Just AI): A spin-off initiative in the Ministry of Justice where engineers embed with prison staff to use AI for security, such as preventing the flow of contraband into prisons.
• Policy Simulation: Tools that allow policy teams to model the impact of decisions (e.g., Universal Credit adjustments) on household finances before implementation.

4. Key Arguments and Perspectives

• Government as an Industry: Mulgrew argues that the civil service should be viewed as a complex industry of 400,000 people ripe for technological disruption.
• The "Outsider" Advantage: Bringing in talent from the private sector is essential because these individuals often stay to build their own internal teams, creating a multiplier effect for technical capability.
• Transparency: The team emphasizes public-facing dashboards (e.g., the AI Opportunities Action Plan) to hold the government accountable for its progress.
• Addressing Sycophancy: When asked about AI models telling ministers what they want to hear, Mulgrew noted that they "red team" models for bias and provide upskilling to users to ensure they understand the limitations of the tools.

5. Notable Quotes

• "We recruit exclusively outsiders... one of the best ways that I can have impact is by getting some people [from the tech industry] into government."
• "You’ve maybe done good stuff in industry, that’s brilliant. Come join us, and we’ll give you the keys to the state and see what you can do."
• "This at the moment is basically a hack to get around the system. So, we need to change that first of all [to make it business as usual]."

6. Synthesis and Conclusion

The 10DS team is successfully demonstrating that small, elite, and highly autonomous technical units can bypass traditional government bottlenecks to deliver rapid, high-value AI solutions. By focusing on both "low-hanging fruit" (workflow automation) and deep-seated systemic issues (planning, criminal justice), they are creating a blueprint for modernizing the civil service. The ultimate goal is to transition these "insurgent" practices into standard operating procedures (BAU) across the broader public sector, moving from isolated experiments to horizontal, scalable AI adoption.