Open source models are really elite for anything high volume
Break your high volume agentic workflows into qwen classifier + claude subagents for less cost higher accuracy
Feels like llms have gotten less creative - maybe an intentional decision to make agentic tasks work
Someones gotta backfill the market for creative uses. An actual creative one, not just randomness / high temperature
something I noticed about bad long-term thinkers are they abstract and hide away from short-term implementation details whereas good thinkers use those small details to guide the entire trajectory of the future
Software engineers, QA, and DevOps eventually merged into one role
With coding agents, it seems software engineers, designers, and product managers are merging into one role
@Axel_bitblaze69 I think @virtuals_io and @autonolas did a lot of foundational work for this 1-2yrs ago. Not enough devs in AI x crypto though
But agreed that crypto is a natural currency for agents! and also for the metaverse
@mattshumer_ AI is amazing in startups, but when you put Opus 4.6 in large codebases (like big tech) they still struggle a lot
On top of that, if you're building anything valuable or new, AI can't do it well because it's not in it's pre-training data / RL task set. Chicken and egg problem