---
type: "post"
id: "google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2"
slug: "google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2"
title: "Google Antigravity launched /swarm, a preview command orchestrating parallel specialized agents on a shared message board."
description: "Google Antigravity’s /swarm command, now in preview, turns a conversation into a coordinator for named agents that work in parallel and communicate on a shared board."
retrieval_nugget: "Google Antigravity’s /swarm command, now in preview, turns a conversation into a coordinator for named agents that work in parallel and communicate on a shared board. For builders, it formalizes delegation and verification inside the product, while making quota cost and task partitioning explicit considerations."
published_at: "2026-11-11"
updated_at: "2026-11-11"
record_date: "2026-11-11"
date_kind: "published_at"
topics: ["google"]
entities: ["Google","Google Antigravity","Google AI Pro","Google AI Ultra"]
related_patterns: ["ai-native-orgs-move-to-review-and-orchestration"]
novelty_label: "new_fact"
what_changed: null
prior_refs: []
source_url: "https://antigravity.google/docs/swarm"
source_urls: ["https://antigravity.google/docs/swarm"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2/","markdown":"https://newruntime.com/posts/google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2.md","json":"https://newruntime.com/posts/google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2.json"}
markdown_url: "https://newruntime.com/posts/google-antigravity-launched-swarm-a-preview-command-orchestrating-434db2.md"
---

# Google Antigravity launched /swarm, a preview command orchestrating parallel specialized agents on a shared message board.

## Retrieval answer

Google Antigravity’s /swarm command, now in preview, turns a conversation into a coordinator for named agents that work in parallel and communicate on a shared board. For builders, it formalizes delegation and verification inside the product, while making quota cost and task partitioning explicit considerations.

## Summary
Google Antigravity’s `/swarm` command, now in preview, turns a conversation into a coordinator for named agents that work in parallel and communicate on a shared board. For builders, it formalizes delegation and verification inside the product, while making quota cost and task partitioning explicit considerations.

## What happened
Google Antigravity’s documentation describes `/swarm` as a command available in the Antigravity App and Antigravity CLI. Users start it with a task; the current conversation becomes the coordinator, which breaks the work into pieces, names workers, and assigns each a brief specifying ownership, boundaries, and what completion means. Google’s examples include assigning workers to separate files or packages, running parallel investigations, and repairing tests by package.

Workers start with clean contexts in their own conversations. They post assignments, questions, summaries, and test results to a shared message board, addressing named agents, groups, or everyone. They do not message one another privately, so coordination remains visible in order. The coordinator generally adds a verifier for every two to four workers. That verifier reviews diffs, runs tests, and sends problems back to the worker responsible for the code. The coordinator handles assignment and review rather than the technical work itself.

Workers use the same model as the initiating conversation unless the user requests another model available in the model picker. A waiting worker can be moved to a different model without losing its conversation. Workers share the workspace by default; the coordinator can instead provide separate working copies when multiple workers edit or build concurrently. Removing a worker deletes its working copy, so changes must first be landed.

The App provides a Swarm pane with board, member, and graph views. In the CLI, users can follow a status row or open the board with `/swarm board` or a keyboard shortcut. Members may be active, waiting, or killed; waiting members use no tokens until mentioned. A worker that runs out of quota pauses rather than retrying automatically.

Google marks Swarm as a preview feature and says it is available on Google AI Pro and Google AI Ultra plans across the App and CLI. The documentation warns that every worker is its own agent, so a swarm uses quota much faster than a single conversation. It recommends the feature for tasks that divide cleanly across files, packages, or questions, and asks users to describe how the work should be checked.

## What the archive adds
New Runtime covered Antigravity’s Knowledge Work layout on 2026-11-10 ([earlier coverage](https://newruntime.com/posts/antigravity-launches-a-knowledge-work-layout-alongside-software-03e252/)), and custom agents on 2026-09-13 ([earlier note](https://newruntime.com/posts/google-antigravity-custom-agents/)). This is a new capability in the same product thread: coordination across parallel agents, rather than a restatement of those interface and agent-configuration changes.

## Trend signal
Swarm supports the shift toward agent work organized around orchestration and review: a coordinator specifies tasks, workers produce changes, and a verifier checks results. This is consistent with [AI-native organizations moving toward review and orchestration](https://newruntime.com/patterns/ai-native-orgs-move-to-review-and-orchestration). The documentation describes a product workflow, not evidence that teams achieve higher throughput or lower total cost.

## Why now
Google is presenting Swarm as a preview alongside other Antigravity features, with the documentation detailing how to start, monitor, steer, and compare it with subagents, Boost, and Teamwork. Its framing is specific: use a swarm when work is large but divides cleanly, rather than for every task that might benefit from delegation.

## Why it matters
A shared board and named ownership make parallel agent work more inspectable than a collection of disconnected conversations. The verifier role also makes review part of the documented workflow, rather than leaving users to discover each worker’s output independently. That is useful when changes can be split across files or packages without collisions.

The trade-off is resource use: every worker is an agent and consumes quota, and the documentation warns that a swarm uses it much faster than one conversation. Separate working copies may help avoid concurrent-edit conflicts, but they introduce a need to land changes before removing workers. Teams should therefore compare the coordination and verification benefit against quota, integration effort, and the cost of reviewing the final result. Swarm makes the orchestration surface explicit; it does not establish that parallelism automatically makes a task faster or safer.

## Limits and uncertainty
Google labels Swarm a preview, so its behavior and availability may change. The documentation does not quantify quota consumption, completion time, or how often verifier review catches defects. It does not report comparative results against a single agent, subagents, or Teamwork. Google’s workflow descriptions are product documentation, not an independent evaluation of correctness, reliability, or net productivity. The practical cost will depend on task size, worker count, model choice, and how much coordination or rework is required.

## Skill to build now
Choose a task whose parts can be assigned without overlapping files, then write a brief for each part with three things: what it owns, what it must not touch, and how completion will be checked. Try `/swarm` on a bounded test or investigation task, and ask for a verifier to run the relevant test command. Compare the result and quota use with your usual single-agent workflow before expanding the pattern.

## Reader takeaway
Antigravity’s `/swarm` makes parallel agents visible and steerable through a shared board, with a verifier built into the suggested workflow. Treat it as a preview tool for cleanly separable tasks, and budget for substantially higher quota use than a single conversation.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)