What happens to software as AI becomes more powerful?

As knowledge computation declines in cost towards $0 and foundational models become more powerful, whither software companies?

A quick post about an idea I’m puzzling over. It’s hard to know what to make of software investment right now—especially if you assume that AI will eat a lot of software companies over the next decade or so. Every so often I get inquiries from people outside of technology about whether they should invest in a particular tech startup. My answer used to be something along the lines of, “If you can afford to lose the money, and you like the team they’ve assembled, sure, why not, lottery tickets don’t print themselves.”

But for any given software startup at this point, we have to start asking the question: to what extent will this company’s software be instantiated in future, more powerful, AI models? I don’t know the answer to that question, and the people who ask me for advice certainly don’t. As the cost of intelligence computation declines to $0, it seems possible that a lot of software companies will disappear.

Maybe the ones that survive are the ones that build on top of commoditized large language models like GPT4 or its successors. I’m not talking about so-called GPT-wrappers that merely encapsulate GPT functionality in a pretty UX. Rather, I’m talking about full-featured, sophisticated software products which use LLMs’ intelligence computation as a substrate. I don’t have a great thesis on what this kind of company looks like.