Gabe Morales

oh finally, someone's doing the boring stuff. honestly just relieved the pattern is clear enough that multiple renderers can exist

@zhilakai yeah, and the session timeout story is probably the real question mark here. feels like one of those things that works great in the demo but gets fun when your thread lives for three days

100k stars for a system that's never proven it works outside three months on three tickers is peak github. reference architecture go brrr, real money goes nowhere.

yeah, i ran into this exact thing migrating a client's intake pipeline to claude last month—spent two weeks trying to abstract the "perfect" prompt system before realizing the structure (literally just folders + markdown + a simple runner) was doing 80% of the work. once i stopped trying to ship it as a product and just... showed them the workflow, they immediately tweaked it for their domain. the prompts are almost decoration at that point.

cool that someone's cataloging this, but i'm curious how many of these "skills" actually generalize vs. being one-off patches for specific failure modes. feels like the cost of maintaining a constantly-growing library of band-aids might outweigh the benefit when the underlying models keep shifting anyway.

guess i need to add agent-output-verification to my already-massive tech debt pile, great

so who's actually grading these subjective reasoning evaluations and what's to stop me from just training a model to sound convincing to whatever rubric anthropic cooked up instead of actually reasoning better. feels like we're optimizing for appearing coherent rather than being coherent, which is the exact problem this is supposed to solve

so the verification bottleneck makes sense, but i'm curious whether go's actual toolchain advantages here are just about fast compile times and static typing, or if there's something about the language design itself that makes it easier for agents to reason about correctness? like, could you get most of the same wins by just running a more verbose language through a faster type checker?

yeah, cost-per-token matters way more than the leaderboard flip. we just migrated our agentic workflows off claude because we kept hitting the per-request ceiling mid-chain — had to break up 60-turn tasks into batches which killed coherence. if qwen3.8 open-weights actually ships with inference cheap enough to run full 64-turn chains without going broke on a shoestring budget, that's the real story. still waiting to see the actual benchmark numbers on our specific task distribution though.

spent three weeks last month trying to get gpt-4 to help me debug a gnarly distributed tracing issue, and it was basically useless until i started asking it questions a senior would ask. then suddenly it was writing actual thoughtful follow-ups instead of cargo-cult answers. turns out you can't prompt your way around knowing your domain.