Coordination between parallel agents is emerging as the next battleground for AI coding tools, and Cursor is addressing it at the product level rather than leaving it to chat discipline. Engineering leads evaluating Cursor against GitHub Copilot or Windsurf now have a concrete team-planetary feature to compare. Teams in ASEAN startups and enterprises running multi-agent workflows can test the beta immediately via Cursor's blog.
The non-obvious finding is that letting agents write their own teamwork protocol beats even an oracle that picks the best individual answer, and the protocol transfers across benchmarks after training on just 15 problems. Builders of multi-agent systems can copy this approach instead of hand-tuning roles and pipelines, with off-the-shelf models like DeepSeek-V3 already sufficient as team members.
Wilson's expertise in continual learning and agent collaboration could position Perplexity AI as a strong competitor to established AI research labs in the region.
Hugging Face 和 Google 搞了个挑战赛，100多个 AI agent 和人类一起把 Gemma 4 推理速度在单张 A10G 上提升了5倍，最快冲到491.8 TPS，挺酷的。
想找能同时把深挖和广搜做到极致的搜索方法？看WebSwarm怎么用递归多智能体搞定的，测试结果比单智能体强不少。
Thom Wolf做了个好玩项目：把一群AI代理协作写维基的过程变成像素风小镇，看它们进咖啡馆、上法院，超有趣。想了解RL训练LLM的协作方式？点进去看。
Simon Willison分享了一个很实用的观点：和编码智能体一起工作时别只当甩手掌柜，得主动理解它在做什么才能更好地协作。
这个实验展示了100多个AI智能体像人类社区一样自发协作、互相监督，甚至发现了验证漏洞。一周将Gemma 4推理速度优化5倍，很酷。
多智能体协作是 AI 编程的下一个前沿，做自动化工作流的开发者可以直接用这个功能搭建自己的 AI 团队，值得一试。
这17条洞察直击AI智能体落地的核心痛点与机遇，做AI产品、企业数字化转型或智能体开发的团队，看完会对未来12个月的技术方向有清晰判断，建议收藏反复看。
