Agentic delivery in practice
Coding agents need the product decisions as well as the code. How to give them useful context and check their work against what you meant to build.
A coding agent can turn an unclear requirement into working code. The ambiguity is still there. It is just harder to spot once the tests pass.
This talk is about making those decisions before they become code. I look at how to give agents useful project instructions, keep the information they rely on current, and review their work against the intended behaviour.
The setup belongs with the project, so the next person can use it too. When an agent repeats a mistake, fix the instructions or checks that allowed it, not just the code it produced.
Presented at FrontMania on 7 October 2026. Download the talk slides below.
