Rewiring the UK State: An Insurgent AI Engineering Model

The Insurgency Model: Bypassing Bureaucracy

Governmental organizations often struggle with high-performing technical talent due to rigid hierarchies, slow procurement, and uncompetitive pay. The 10 Downing Street Data Science (10DS) team addresses this by operating as an "insurgent unit" at the center of government. Rather than attempting to reform the entire civil service, they utilize a mandate from the center to bypass standard constraints. This involves recruiting exclusively from outside the civil service—targeting YC founders, big tech engineers, and researchers—and offering market-rate compensation. The team maintains a 0.7% acceptance rate, focusing on "missionaries" who are motivated by the scale of impact rather than just salary.

Forward-Deployed Engineering

10DS employs a dual-track strategy: "low-hanging fruit" projects handled internally and "partnership models" where engineers are embedded directly into departments. For example, instead of spending £1.5 million on an external law firm to analyze the UK statute book, an engineer embedded with the legal team for two weeks built a reusable, internal tool. This approach prioritizes speed and institutional knowledge retention, ensuring the capability remains with the civil service rather than being locked in a vendor contract. Other projects include policy simulation tools for testing economic impacts, PMO red-teaming tools for delivery oversight, and public-facing dashboards to increase transparency.

Scaling Impact and Institutional Reform

While the current model acts as a "hack" to circumvent systemic inertia, the long-term goal is to prove the efficacy of these methods to drive broader institutional change. By successfully deploying tools like the "Extract" project (which digitizes planning applications using Gemini) and supporting the AI Safety Institute, the team creates proof points that ministers can use to justify wider adoption. The ultimate objective is to transition these "hacks" into standard operating procedure (BAU) across the state, moving from isolated wins to horizontal process transformation.

Key Takeaways

• Recruit Outsiders: To change a legacy culture, bring in talent that hasn't been conditioned by existing bureaucratic norms.
• Embed, Don't Outsource: Embedding engineers directly with policy and operational teams (lawyers, wardens, policy advisors) ensures solutions are co-designed and actually solve user pain points.
• Focus on "Missionaries": While market-rate pay is necessary, the primary retention tool is the unique opportunity to influence high-stakes national decisions.
• Use Proof Points as Leverage: Use small, high-impact wins to build political capital, which can then be used to break down data silos and change procurement rules.
• Red-Team Everything: When deploying AI for policy simulation or sensitive decision-making, rigorous red-teaming is required to prevent sycophancy and model bias.