From coding with more comprehensive autocomplete to developing with agentic vibes, developers, teams and companies have adopted LLMs into their workflows. Many, however, are doing so without a clear sense of the trade-offs involved or full awareness of whether they are receiving meaningful benefits or just the benefit of illusion, to the point some developers seem to be rapidly deskilling themselves, turning themselves into fixers of newly created legacy, firefighting their way through thickets and tickets of busywork.

We are often told that as long as we keep a "human in the loop" it's all good. At one level, this advice seems sound — and even obvious — but it's as vague as it is unhelpful. What does it actually mean to be the human in the loop? It's clearly more than just LGTMing any generated code, but what skills and mindset are needed to make AI use effective? What does it take to be the driver, not the road? How do we avoid cognitive surrender, skills atrophy and burnout?

This isn't a talk about tools, agent herding, better prompts or markdown style tips. It's about understanding what productivity is and is not, the difference between value demand and failure demand, the mechanisms behind skills and learning, cognitive offloading versus cognitive surrender, human connection and collaboration, and the value of generating options over linear solutioneering.