Embracing agentic development unleashes an army of eager, expert amnesiacs on your project. Agents are well-trained and desperate to please, but they need coaching to produce usable code. Unfortunately, when coaching agents, conversations evaporate. Coaching only counts when it lives in the repo.
In this session we will look at several approaches for successfully directing your agents and making your instructions stick:
- Enforcing architecture with ArchUnit (and its Python and JavaScript ports)
- Continually refining Skills without bloating your context
- Cross-model code reviews
- Docker Sandboxes and YOLO mode
You will leave this session ready to (safely) turn your agents loose while you grab a sandwich.