Defra builds civil-service manual using agentic AI

TL;DR:

• The Department for Environment, Food and Rural Affairs has published a new digital service manual built with the help of autonomous AI coding agents working alongside human designers and researchers.
• The manual replaces guidance previously scattered across SharePoint sites, intranets and disconnected documents — addressing a long-standing complaint from civil-service teams.
• Resultsense view: this is one of the first concrete UK central-government deployments of agentic AI in user-facing tooling reported in 2026 — a small but meaningful proof-point against the perception that Whitehall AI work is stuck in pilots.

Defra’s AI user-centred design lead Chris Leo described the build to UKAuthority as a true collaboration in which AI agents took high-level user needs, chose technical approaches and iteratively refined the code, while content designers and researchers maintained editorial direction and validated decisions through usability testing.

Why this is a step change

Most UK central-government AI announcements over the last 18 months have been frameworks, principles or pilots rather than shipped tools. The Defra build is one of the rarer cases where a department has used agentic AI to deliver a working internal product end-to-end. The team paired AI-driven engineering with conventional GOV.UK content standards and accessibility checks, reflecting the dual track now expected in Whitehall: AI for delivery speed, GDS-style craft for quality.

Subject-matter input came from across Defra’s design, user research, content and accessibility communities, with each section reviewed by professionals in those specialisms. That governance pattern — community contribution plus AI execution — is closer to how mature open-source projects work than to the integrator-led model that has dominated central-government IT for two decades.

UK context

Defra’s move sits alongside a slow drift in central-government tooling. The Government Digital Service has long maintained the public-facing service manual at GOV.UK; departmental equivalents have struggled to keep pace. AI-assisted authoring shifts the maintenance economics: instead of one team owning a static manual that decays, agentic systems can consolidate updated guidance from multiple sources on demand.

For UK businesses considering similar in-house manuals — onboarding, compliance, internal HR — Defra’s pattern is a useful reference. Content designers Anita Shargall and Melissa Massey handled language and standards; researcher Sophie Werkshagen led usability validation. The team have signalled they will pursue further agentic-design work, suggesting more departmental tooling could follow the same path.

Looking forward

The UK public sector has spent years debating governance frameworks for AI without much shipped product. Defra’s manual is a modest counter-example: a working tool, built quickly, with civil-service editorial control intact. Whether other departments adopt the same blend of agentic build and human editorial review will indicate whether Whitehall is ready to apply AI to delivery, not just policy.