Use AI patterns
See what other Defra teams have built with AI.
Each pattern is an early proof of concept built and shared by a Defra team. Treat it as a starting point to adapt for your own delivery, not a finished product.
Find a pattern that fits
Try Agent swarms
Multiple specialist agents working together tend to produce richer analysis than a single agent, especially on complex policy or strategic questions.
Read the agent swarms proof of conceptOr email the team to discuss this for your project.
Try the AI assistant
The team's prototype Defra-hosted AI stack lets you run multiple providers behind a single interface, with consistent guardrails and monitoring.
Read the AI assistant proof of conceptOr email the team to discuss this for your project.
Try Green summarisation
Smaller, local models can match cloud LLMs for short summaries. Useful when sustainability matters and the task is constrained.
Read the green summarisation proof of conceptOr email the team to discuss this for your project.
Try Token optimisation
Prompt compression techniques can reduce token cost on repeat prompts. Results vary by task; the proof of concept shows where it works and where it does not.
Read the token optimisation proof of conceptOr email the team to discuss this for your project.
Try the Interaction designer plugin
A Claude Code plugin that wraps a guided arc and six leaf skills around a Defra designer's workflow. The architecture transfers to other roles too.
Read the interaction designer plugin patternOr email the team to discuss this for your project.
Talk to the team
If none of these fits, your work might be a candidate for a new pattern. Email AICE to discuss.
Email the teamContribute a pattern
If your team has a reusable approach you want to share, AICE can help you write it up.