Bron Afon: AI Good Practice - Set up properly, then earn the autonomy

The practical discipline of doing AI well. Data first — AI built on unreliable data fails. Investing properly in setup: configuration, testing and context-building rather than plug-and-play. Guardrails and rules built in by design, before deployment. Then the maturity journey: from simple prompting, to AI-assisted workflows, to agentic teams with genuine autonomy — always with humans in the loop. Example from our development workflows: AI does the first code review, but a human always checks the code, and no database schema change goes through without human sign-off.

We seek to help answer the following questions

• What's the biggest mistake organisations make with AI?
• What does "agentic" mean and should housing associations care?
• If AI is doing the work, why keep humans in the loop?