Software engineering has been rising in abstraction for at least, say, the last 50 years.
We moved from assembly to C, from C to Java, from SQL to ORMs, and from managing our own infrastructure to AWS.
And this was great. It unlocked new markets and made us more productive. You had a small percentage of real hardcore computer science researchers moving the field forward, and a vast number of commercial software developers using high-level tools to ship things. I benefited massively from that.
But given the refactoring of our occupation over the last two years, driven by agentic coding, I’m betting the next phase of my career on getting closer to the metal.
Or perhaps more accurately: getting closer to the foundations.
The software developer role is already changing faster than at any point in my career. The truth is Claude Code and Codex are already better programmers than most of us. Karpathy warned us in 2017:
For decades, the leverage came from moving up the stack. But it turns out you’re always safe if you’re strong and can focus back on the fundamentals. A strong mathematical foundation. Then enough machine learning and AI knowledge to reason about these systems from first principles. Then moving toward the baseline of a good ML/AI researcher. And optionally going deeper into spatial computing, robotics, world models, or whatever comes next. This doesn’t mean every software engineer needs to become a researcher. But it feels a bit like the vibes I got while reading The Dream Machine. That sense that the personal computer and the internet were not simply products waiting to be invented, but entire decades of possibility slowly coming into view.
And to be part of it, you can get closer to the metal, try harder things and who knows… Maybe build parts of it.