AI-Driven Development

AI doesn't change what's hard about building software. It makes what's hard more visible -- and it introduces new failure modes that map onto decades of safety science.

These articles explore that territory. The first group establishes the fundamentals: what AI changes about the developer's role, where the new risks live, and how to maintain control over an optimization system that will happily optimize for the wrong thing. The second group applies those ideas to real incidents -- what happens when systems drift toward failure and nobody notices until the boundary is crossed. The third examines the cognitive mechanisms that make AI-assisted work subtly dangerous even when it feels productive.

The through-line: the most effective response to AI in development is not to use it better, but to understand what it does to your cognition and design your workflow around that reality.

The observations come from twenty years of professional experience in DevOps and platform engineering. The theoretical frameworks come from graduate coursework in human factors -- safety science, cognitive psychology, and systems thinking. The synthesis is mine: connecting individually-studied cognitive phenomena to the specific context of AI-assisted development. Where a claim is established in the literature, I cite it. Where it's my own inference, the text makes that clear.