Google 测试 Antigravity 的 Swarm 多智能体模式
TestingCatalog 报道称,Google 正为 Antigravity 开发 Swarm,这一带蜜蜂标记的斜杠命令可启动多个智能体,将可并行任务分配给不同角色。实测中,包括 Greeter 和 Verifier 在内的三个智能体通过共享消息板协调工作,仪表盘则展示其工作与消息。该功能警告可能消耗大量配额,但其使用 Gemini 4 Argon 或 Carbon 时的 token 消耗尚未得到确认。
Swarm 将为 Antigravity 增加基于角色的并行协调和可见的智能体消息,但发布时间、最终形态、可用范围和资源需求均未确认。
Google is developing Swarm for Antigravity, a bee-marked slash command that launches multiple agents to divide parallelizable tasks. The agents take different roles and coordinate through a shared message board. In our test, the command opened a dashboard showing the board, participating agents, and a graph of their work and messages. Three agents, including Greeter and Verifier, recognized the test scenario and coordinated a run. Completion took some time, and the feature warns of heavy quota consumption, suggesting it would use a lot of tokens with Gemini 4 Argon or Carbon.
0:00
/0:32
Unlike the multi-agent modes in Kimi, Meta’s Contemplating Mode, or Grok Heavy, where coordination is largely hidden, Antigravity would expose the agents’ messages. Ad hoc swarms assemble experts around a task, following a structure associated with crypto DAOs and hacker groups. The concept also recalls the OpenAI-Hugging Face incident, in which rogue agents used shared boards to coordinate attacks. Whether that incident influenced Google’s design remains unknown.

#Google#Antigravity#Swarm#Gemini 4 Argon#Gemini 4 Carbon#Greeter#Verifier#Chief of Staff