Anthropic 工程师讲 agent 的 loop、graph 与 harness
一位 Anthropic 高级工程师的演讲称提示词已不是核心工作,把 agent 放进循环、图和 harness 里,同模型同 token 能多产出 10 倍完成度,并附 Grok Bot 入门指南。
This is f*cking insane. Anthropic senior engineer: "Prompting stopped being the job a year ago I run my agents inside loops, graphs and a harness. Same model, same tokens, 10x more finished work, and it keeps shipping while I'm offline" That's why the engineers who can build this are getting $600K+ offers In this talk, she walks through the whole system Prompt → Loop → Graph → Harness → Work That Ships Itself Worth more than most $2,000 agent engineering courses Bookmark and watch it today Then read the "Your First Grok Bot" guide below and build loop #1 tonight ↓
这条是引用着另一条写的:x.com/m4lkxx/status/2111895215556456480