Software Has No Ceiling

Is it still worth doing software now that AI can spin it up "so easily." I put easily in quotes on purpose, because the premise is, well, shaky. The part that is interesting to me is what some engineers are doing with that premise. For example, some may have decided that since machines can write code, the smart move is to pivot into hardware. Hardware is harder, so it is safe.

This is one of the most misinformed ideas I have heard in a while.

Hardware is software now. Nobody places transistors by hand or draws a chip with a pencil. The whole stack runs on software. If you are running away from code, hardware is the last place to hide. And AI will likely get good at hardware faster, because hardware is bound by the laws of physics. There is a floor and a ceiling. You are arranging atoms to do something useful, and there is a limit to how cleverly you can arrange them.

Software has no such floor. It lives entirely in a virtual space, a pure concept with nothing physical pushing back. Look at the last decade. Software grew exponentially more complicated, to the point where whole industries exist only to manage that complexity. Hardware got better over the same period too, but it did not run away like that. Physics will not let it.

So which of these keeps scaling upward? The answer is software, every time. The more AI you deploy, the more software there is, and the more tangled the systems around it.

Yes, you can build a todo app or a fitness tracker in an afternoon now. That used to take weeks, maybe months. But those are the easy base cases. Real systems get vicious at scale, and every new line written makes the whole thing harder. Ask a reverse engineer. A small binary is a warm-up. An entire OS is a career - a bit shorter now thanks to LLMs, but still a mountain.

More software gets written now, and a growing share of it is bespoke, which only deepens the mess.

You cannot escape AI by going low-level. There is no low enough.