seriously, working with AI is MISERABLE for one and only one reason: having to re-explain the same thing "oh yeah this new session obviously doesn't know what proper case trees are, so let me explain it for the 5000th time in my life" I'm tired AGENTS.md doesn't solve this because it is impossible to fit the entire domain knowledge without nuking the context - it would be 1m+ tokens worth RAGs don't solve this, the agent won't search unknown unknowns SKILLs don't solve this unless I keep like a collection of 1750 skills with specific cuts of domain knowledge for each possible subset of my domain that I might need in a given chat, but that's a lot of manual work recursive LLMs or whatever don't solve this for the same reason, you can't dump a domain book and expect the AGENT will magically guess that it is supposed to search for a specific bit knowledge. unknown unknowns fine tuning doesn't solve this (OSS models suck and OpenAI / Anthropic gave up on user fine tuning) I honestly think a good product around fine tuning on your domain would be a major hit and an underdog lab should take this opportunity

Maybe progressive disclosure would help with this? I'm not sure what shape it would take, or if agents would do well with it, just spitballing here. I'm imagining like, an LOD tree for knowledge. And maybe this is just SKILLs with more hierarchy? e.g knowledge.md Don't use BigInt - see [bigint/why.md] bigint/why.md We don't use BigInt in this project because short description. When BigInt seems like the right solution see [bigint/but-why.md] But like you say, the hard part is a) creating this thing in the first place b) keeping it up to date c) at some point knowledge.md will still get too big Agents can probably do (a) and (b) but idk how well they'd actually do it and (c) is still eventually an issue. I guess it's kinda just handrolled compression at the end of the day. Maybe train the model on .zip-formatted data and we can just upload our compressed markdown blobs 😂

@dceddia @VictorTaelin I've been toying around with various kinds of fractal decomposition such that relevant complexity is discoverable in context just-in-time, but I'm tired boss