AI and the Future of Programming
The tools are changing faster than the discourse around them. What does programming actually look like when AI is embedded in the loop?
Something has shifted in how software gets written, and I don't think the industry has caught up to it yet. Not the hype — the actual practical reality of what it means to write code today versus two years ago.
The loop has changed
The inner loop of programming — the cycle of write, run, debug, repeat — used to be bounded by how fast you could type and think. Now there's a third party in the loop, one that can generate large amounts of plausible code very quickly.
This changes the bottleneck. The constraint is no longer producing code. It's evaluating code, understanding it, and steering the system toward what you actually want.
That's a meaningfully different skill set than what we've been optimizing for.
What doesn't change
Understanding matters more, not less. When AI can generate ten implementations of a function, the question becomes: which one is right? Which one will scale? Which one will be maintainable in six months?
Answering those questions requires a deeper model of the system than ever before — because you're evaluating outputs rather than producing them from scratch.
The "you don't need to understand how it works" crowd is exactly backwards.
The energy angle
There's an underappreciated infrastructure story here too. AI inference is compute-intensive, and the compute-per-token curves are just starting to bend. The energy cost of the AI tools embedded in your editor is real. The data center build-out required to support this is massive.
These are connected problems. The future of programming is also, in part, an energy question.
More on that in future posts.