Summary
William Tarr and Louis Hogarth from the UK Ministry of Justice’s Justice AI unit describe what it looks like to actually ship AI products inside government — not pilots and slide decks, but tools deployed in prisons, probation offices, and courts that front-line staff use every day. The unit, which grew from one person to roughly 40 in about 18 months, operates on a forward-deployed engineering model: builders embed directly with prison officers, watch them work, and ship MVPs in days rather than quarters.
The talk opens with the crisis that framed the unit’s founding — a 2023 wave of headlines about prisoners being incorrectly released from Wandsworth prison, caused by a combination of disjointed identification systems, broken inter-agency communication, and decades of layered legacy technology with unreliable APIs. Tarr and Hogarth explain how the traditional centralized government tech model — where builders rarely interact with the people using their software — produced this kind of failure, and how their unit is structured to close that gap.
The second half of the talk addresses organizational friction: resistance from HQ staff who have added ‘AI’ to their job titles without hands-on experience, union engagement, ethics and privacy review processes, and the challenge of navigating procurement without losing momentum. Their tactic — bringing real pilots with early signal to governance conversations rather than theoretical proposals — has allowed them to move faster than typical government timelines while maintaining safety and security standards from day one.
📺 Source: AI Engineer · Published July 08, 2026
🏷️ Format: Workflow Case Study
