Two engineers build and ship UK prison software in weeks

Traditional government technology projects fail because developers build them from distant desks. William Tarr and Louis Hogarth from the Justice AI Unit—an agile team inside the 95,000-person UK Ministry of Justice—rejected this legacy approach. Instead, they adopted a forward-deployed engineering strategy. Engineers do not just write code; they walk prison wings, hold physical keys, and sit beside probation officers.

By working directly on-site, builders observe the stark reality of daily operations. They bypass the idealized procedures written in headquarters manuals, focusing instead on how officers actually behave. This immediate contact allows the unit to build minimal viable products in days rather than quarters.

Coding for Victorian Walls and Zero Wi-Fi

Building software for prisons introduces harsh physical constraints. Many UK prisons are Victorian-era structures built with seven-foot-thick stone walls. Wireless signals cannot penetrate these barriers, and security policies often ban Wi-Fi altogether. Tech teams in headquarters regularly roll out digital tools that are instantly useless because they require constant connectivity.

To solve this, the unit writes applications with robust offline modes, caching data locally in the browser. When probation officers encounter spotty cellular connections in remote offices, the software continues to work. The team also prioritizes extreme simplicity. By keeping interface design basic, they make the tools easy to use, ensuring that busy staff can use them under pressure without extensive training.

Fighting Committees with Working Software

Government procurement processes are notoriously slow. Bureaucratic systems apply policy-based governance to product design, forcing ideas through months of committees staffed by people who have never set foot in a correctional facility. This structure creates heavy risk aversion.

Traditional: 40-person teams -> Months of planning -> Sluggish pilots -> No deployment
AI Unit: 2-person teams -> Rapid prototyping -> On-site deployment -> National rollout

The team bypasses this friction by building fast and demonstrating safe-to-fail pilots on-site. Showing a running application with real-world user data shifts the conversation from theoretical risks to observable benefits. This practical proof gives them the leverage needed to streamline approvals.

The Shift to Lean Public Sector Tech

The success of this model has massive implications for public services. Just two engineers built, tested, and completed a national rollout of a probation tool in a few months—a task that typically employs dozens of bureaucrats. By shifting focus toward automated transcription, predictive intelligence, and streamlined data integration, the unit aims to return officers to face-to-face public service, proving that even the largest bureaucracies can ship production-ready modern software.