But while the AI was ready to give up several times, it did keep adding debug code and analyzing it faithfully when I pushed.
能力其实在，缺的是自主坚持——这是 Linus 这条记录里最有价值的判断。对 agent 设计的启示不是继续堆推理长度，而是加一层"不得自行宣告不可解"的约束。反面推论也成立：不会追问、不敢施压的使用者会系统性地拿到更差的结果。
4 Matching Annotations
- Aug 2026
-
simonwillison.net simonwillison.net
-
- May 2026
-
blog.cloudflare.com blog.cloudflare.com
-
Humans can be in the loop to grant permission and must accept Cloudflare's terms of service, but no human steps are otherwise required from start to finish.
最佳实践建议：自动化流程可以大幅提高效率，但人类审核和接受服务条款仍然是必要的。
-
- Apr 2026
-
-
because coding has a tight human-in-the-loop workflow, with developers still overseeing the development process today, these tools enable accelerated output while still making space for human judgment to review, edit, and iterate.
「人在环路」是编程 AI 爆发的关键因素，而非阻碍。这个洞见颠覆了常见的「人机协作摩擦论」：恰恰是因为开发者需要审查代码，AI 生成的错误有人把关，企业才愿意大规模部署。这说明 AI 在「可验证 + 人类兜底」的领域最容易突破——其他领域想复制这个成功模式，需要先建立同等的验证机制。
-
-
arxiv.org arxiv.org
-
Each task includes a unified evaluation framework supporting sandboxed code and APIs, alongside a human reference trajectory annotated with stepwise checkpoints along dual-axis: S-axis and V-axis.
大多数人认为AI评估可以通过简单的自动化测试完成。但作者提出需要复杂的双轴(S-axis和V-axis)人工参考轨迹和沙箱环境支持，这暗示了评估AI代理能力的极端复杂性远超当前行业的普遍认知。这一观点挑战了AI评估的简化主义倾向，强调了人类参与在评估中的不可替代性。
-