Code is the medium.
Systems are the work.

I'm not interested in clever code. I'm interested in code that a team can modify safely six months after I've shipped it. That means clear boundaries, explicit data contracts, and tests that test behaviour — not implementation details.

The hardest engineering problems I've worked on weren't algorithmic. They were organisational: how do you evolve a live system without breaking the customers using it? How do you make a junior developer productive without creating a dependency on you?

AI has changed what's possible, not what matters. The fundamentals — clear data models, good API design, systems that fail gracefully — are more important now, not less. I use AI deliberately, as a tool, not as a substitute for thinking.

I think in systems, not in sprints. Any engineer can ship features. The discipline is in the invariants you maintain while doing it.