The Sovereign Engineer: Staying in Command of Code You Didn’t Write
Part of the AI Literacy Series
(Available as a keynote and session talk, in-person and virtual,
conference or internal to an organisation)
You reviewed it. You shipped it. And when it breaks at three in the morning, it's yours. But you didn't write it.
The deadline is close. You're tired. The generated code looks reasonable. The tests pass. And, almost without noticing, you stop reading quite so carefully. Weeks later, you're looking at a codebase full of commits with your name on them and realise you can't quite explain why some of it works the way it does.
Peter Naur gave us a useful way to think about this forty years ago. In Programming as Theory Building, he argued that the real product of programming isn't the program text. It's the theory held by the programmers: their working understanding of why the system is built this way, how its parts relate, and how it can be changed safely. For most of our history, code and theory grew together. AI changes that relationship.
We can now acquire code much faster than we acquire the theory behind it. So the interesting question isn't whether we should use AI to write software. We already are. The question is how much understanding we can afford to delegate with it. Which decisions are easily reversible? Which ones will shape the system for years? How much damage can a wrong decision do? And how will we discover that it was wrong — seconds later through a test, during review, or six months later in production?
This talk uses those questions to describe a progression in AI-assisted engineering: from simply using AI, to systematically verifying its work, to deliberately shaping the environment in which human and AI decisions are made. That last step is where sovereign engineering begins. Not refusing delegation, but remaining capable of judgement.
You'll leave with a way to recognise where you are on that progression, and a practical idea of what your next step might be.