There's a way I've been working with AI that I call generative engineering. It's a little different from simply asking an AI to write code. I might start with an idea for a software system, an interface, a scientific simulation, or even a physical engineering problem. I'll work through the architecture with AI, establish what already exists, what needs to happen, and what constraints the solution has to satisfy. Then things get interesting. I use different AI models and agents for different parts of the engineering process. One might help develop an architecture, another look for problems with it, another implement components, and another independently evaluate the results. Sometimes several agents work on different parts of a system in parallel. Sometimes I have AI systems review and improve the engineering instructions that other AI systems will execute. The prompts themselves become engineering artifacts, with defined objectives, constraints, dependencies, acceptance criteria, and boundaries around what can be changed. I don't have to personally write every line of code, but I do need to understand what's being built, how the pieces interact, and what the results actually demonstrate. And things don't always work. We've had seemingly successful improvements introduce regressions elsewhere. We've had simulations produce conflicting results because the underlying experimental inputs weren't actually equivalent. We've had designs that looked promising until verification exposed a fundamental architectural problem. When that happens, we don't just keep generating until something looks right. We investigate, measure, preserve what we've learned, revise the approach, and test again. Sometimes that means changing a component. Sometimes it means reconsidering the entire architecture. We maintain the history of decisions, experiments, failures, and verified results so that the next stage of work builds on what we've actually established, rather than whatever an AI happens to remember or claim. That's an important part of it. The engineering process itself becomes cumulative. What's remarkable to me is how this changes the scale of what an individual engineer can work on. I can move between software development, electronics, simulations, mathematical models, and systems architecture, coordinating generative tools to explore and develop solutions that would otherwise require considerably more time or a multidisciplinary team. And the AI isn't limited to producing the final artifact. It can participate throughout the process, from exploring requirements and developing candidate architectures to implementation, testing, fault analysis, and optimization. The engineer remains responsible for the design intent, the constraints, and determining what evidence is sufficient to move forward. I think of it as moving from manually constructing every component toward orchestrating the engineering of entire systems. That's generative engineering as I practice it. The objective isn't to generate as much code as possible. It's to expand what we can design, construct, understand, verify, and ultimately accomplish through engineering.