Google Antigravity Adds /swarm to Run Teams of Agents in Parallel
Announcing /swarm, a new slash command in Google Antigravity for orchestrating specialized agents in parallel across the desktop app and CLI. Turn a single prompt into a coordinated multi-agent workflow where agents plan, build, and verify together on a live message board, with the flexibility to mix and match different models for each role.
Context
It moves Google's coding tool from one agent at a time to coordinated groups of agents that can each use a different model.
Antigravity is Google's agent based coding environment, available as a desktop app and a command line tool. Google launched its Gemini 4 Carbon model on the same day, and Sundar Pichai shared a demo built with Carbon and the new /swarm feature.
Deep dive
Google Antigravity announced /swarm, a new slash command for orchestrating specialized agents in parallel across the Antigravity desktop app and its command line tool.
According to the post, /swarm turns a single prompt into a coordinated multi agent workflow in which agents plan, build and verify together on a live message board. Google said users can mix and match different models for each role. The short video attached to the post shows the Antigravity prompt box with a menu of slash commands and a model picker set to Gemini 4 Argon Medium.
Earlier the same day, Sundar Pichai posted a demo he described as a fun test of long horizon coding and the new /swarm feature, built with Gemini 4 Carbon. His post drew about 250,000 views, more than the Antigravity announcement itself.
The announcement does not give pricing, usage limits, or the maximum number of agents a swarm can run, and Google did not publish benchmark results for the feature.
What to expect next
Watch for developers to share how /swarm handles larger codebases, how much a multi agent run costs compared with a single agent, and how reliably the verifying agents catch mistakes. Parallel agent teams are becoming a common feature across coding tools, so similar launches and comparisons from Cursor, OpenAI and Anthropic are likely to follow. Google may also publish guidance on which models work best for planning, building and verifying roles.