Comment by dewey
6 days ago
Maybe I'm missing something, but wouldn't using the most popular open source framework + most popular open source database be a much better fit as there's much more training data and you don't lock yourself into yet another framework?
I'd just go with Rails + Postgres and have all the documentation and options open to me.
This is exactly it, and people trying to invent these new tools and formats (wasn’t there a “LLM optimized JSON alternative the other day) are seriously not understanding how LLMs are trained a tremendous amount on understanding JSON and XML and standard tools and frameworks.
I disagree. The LLM seems extreme adaptable to me. Making the architecture that makes the most sense, that expresses the architecture the best, is going to be a soaring victory over bandwagoning every time.
It's not at all about picking popular tech: that's what the limitation used to be. Now it's all about finding the best winds to have at your back.
Reducing your app code complexity, having a better defined more cohesive base to build upon will reduce reasoning costs & token count. It will also make it easier to assess what you+agent have built too.