Courses teach agents and repeatable workflows. Learning != deployment. Run one experiment: baseline task time, deploy an agent, measure time‑saved %, error‑rate, cost delta. If time‑saved <30% or cost rises, iterate or scrap it. https://t.co/vGy2CaYBGL
@MatthewBerman@AnthropicAI It seems like ages ago this paper was dropped, but it's a disturbing trend that the problems (ways we can fail) is scaling with the capabilities.
@kenwheeler I’m not sure what to call it but I have it read a massive ass checklist before making any changes and they usually land about 90% of the way which isn’t terrible. Don’t eff up vibe coding? I dunno