ro-han_solo
Joined 6 years ago
Joined 6 years ago
it chose plain solid wo a frameworks since i wanted the backend in rust anyway.
the rationale was streaming ai responses is better in it, since react re-diffs the dom every time a token is streamed, while solid just updates the exact node text changed.
tbh nowadays ive stopped caring about frameworks. wtver is comfier for the agent to write in, with enough back pressure from tests, ci and lifting so that agent fuckups dont reach me is good enough for me
this is what i was building https://github.com/clauseo-ai/boxbox2
this plus pdf parsing and structured output extraction using llms is pretty much mostly what I do
what got me was finding out how much time went into making the model comfortable in my harness. it kept reaching for tool names and arg shapes it was trained on, so I ended up writing shims that quietly translate them into mine.
atp im basically rebuilding their harness w extra steps lol.
the stuff thats actually mine is model choice, workflow and annoyances like cleaning up after themselves, along with thread to thread communication which i really like, and also just knowing every aspect of a harness.
I also hacked a mechanical numpad to be kind of a codex micro lol, need the client to be mine for that.
cmx2 - https://github.com/clauseo-ai/cmx2 boxbox - https://github.com/clauseo-ai/boxbox2
models are being heavily rl'd on the lab's harnesses, so eventually claude works better in claude code than any other harness.
i'm moving my weird customisations a layer above to acp or building on the claude sdk, especially since they now allow using your sub w it.
the labs are making it clear that they want control of the harness as well
we should be prepared for that. a bunch of enterprises were shocked after github switched to usage based pricing for copilot lol, and now devs stuck w just $29 of spend for a month are just coping.
i think as long as the economic benefit to a person is higher than their inference cost, it's fine, but a lot of the wasteful burn of tokens on silly things will most likely end
longer running threads + cron jobs and more diverse tool selections would pbv lead users to use up more tokens.
makes me curious about instinct's cashflow more tbh, they're free right?
hermes and openclaw burn a lot of tokens for me lol