To do this, I've developed a process for working with coding agents, particularly Claude Code, which allows me to switch context rapidly and do things I don't 100% know how to do myself.

In this talk I'll share the techniques I've learned, developed and refined that help me get solid, usable code that I can put into production with confidence. I'll show you how to write a detailed requirements document that an agent can work from; how to work through those requirements step by step in collaboration with your LLM; how to maintain state between sessions (which is helpful for me as well as Claude); and how to keep up with what it's doing so you can take responsibility for the solution long after its context window has closed forever.

I'll also share some thoughts about where junior developers fit into this new way of working, and how important it is that humans continue to be the primary agents in software engineering.