July 9, 2025
"What was the last platform shift like, Grandpa?"
The GPT-4 release was my wake-up moment. The GPT-3.5 to GPT-4 improvement convinced me that software was going to change in a big way. (Other people saw this earlier than I did. Credit to them!)
But even so, I thought it wasn't clear exactly how big the change was going to be, or what the best historical analogies were. The relational database? The Internet? The steam engine? Perl-compatible regular expressions? 😇
I wrote an internal strategy memo for @trydaily that said:
1. the relational database is the lower-bound for how important this new LLM stuff is going to be. LLMs will be a fundamental component of almost all production software.
2. The upper bound is the Internet. What software can do, how we distribute it, and how we write code may all transform massively.
Two years later it's pretty clear we're in upper bound territory. This is a *big* platform shift.
But, interestingly, I still think the analogy to relational databases is helpful in thinking about how to plan for an LLM-centric future of software.
fun intellectual exercise: replace "LLM" with "database"
"local databases are the future."
"there will be a duopoly of frontier LLM labs"
"databases are the fastest commodifying product in human history"
last one was probably true back in the day! the other two? idk.