How fast can we go?

I’ve been thinking a lot, lately, about the now-ubiquitous tech industry trend toward LLM automation of coding and the expected and anticipated productivity gains that are touted as reasonable and achievable. Every big name you can think of is driving AI adoption, not just within their products (where, there too, it’s ubiquitous), but also in their operations. The theory goes that individual ICs can be made dramatically more productive through the adoption of LLMs in their day-to-day work. In these statements I’ve seen floated what felt to me like unrealistic claims about the level of acceleration that might be possible. We’re not talking 50% or 100% gains in productivity (which would be incredible). We’re talking 1000%, 5000%. Pick any number you think is reasonable. Now 10x that. But I realized, as I was thinking about this, that my own reaction is based on feelings, not facts. So I figured, fuck this, let’s see how far I can get with some basic modeling that might help me ground…