---
title: 'Google Antigravity Just Got Swarm Mode (9 AI Agents, 1 Prompt)'
source: 'https://youtube.com/watch?v=ZQscIaVi_9g'
video_id: 'ZQscIaVi_9g'
date: 2026-11-15
duration_sec: 448
channel: 'Julian Goldie SEO'
---

# Google Antigravity Just Got Swarm Mode (9 AI Agents, 1 Prompt)

> Source: [Google Antigravity Just Got Swarm Mode (9 AI Agents, 1 Prompt)](https://youtube.com/watch?v=ZQscIaVi_9g)

## Summary

Google's Antigravity platform now integrates Gemini 4 Carbon and a new 'Slash Swarm' mode, enabling users to spawn multiple AI agents that work in parallel on a single project. The video demonstrates building three real business applications using this swarm approach, with verifier agents ensuring quality. It highlights the practical setup, key learnings, and how to replicate the process.

### Key Points

- **Introduction to Gemini 4 Carbon and Slash Swarm** [00:00] — Google Antigravity now includes Gemini 4 Carbon, their strongest model for long software jobs, and Slash Swarm, a new mode that allows a chat to act as a coordinator, spawning parallel workers with isolated project copies.
- **Verifier Agents and Swarm Pane** [01:02] — Verifier agents challenge the work by running tests and opening the app, sending broken items back. The swarm pane shows a live board, members list, and a graph to scrub through the run.
- **Demo 1: Client System for Web Design Agency** [01:44] — A prompt for 'client comb' builds a client system with four workers (leads, follow-ups, bookings, reviews) and a shared shell. Verifiers check in a real browser, passing with zero console errors and zero overlaps in 24 minutes, producing 11 files.
- **Demo 2: AI Audit Kit for Automation Consultant** [03:24] — The 'audit comb' uses four workers (intake, process map, scorecard, report) to build an AI audit kit. It generates 477 KB of code in 32 minutes, finding 48.5 hours a week of automation potential for the demo agency.
- **Demo 3: Weekly Update Desk for Coaches** [04:33] — The 'keeping up comb' is a weekly update desk for a coach. Four workers (inbox, triage, actions, member post) process real updates. Verifiers caught issues, sent workers back, and after fixes, passed 57/57 checks with zero errors.
- **Key Learnings from Running Swarms** [05:29] — Be specific about ownership (one module per worker), write verify rules into the prompt (real browser, no console errors, no overlaps), be mindful of quota (each worker is an agent), and use the graph tab to review the run.
- **When to Use Swarm and How to Try It** [06:10] — Swarm is ideal for jobs that split into 3-4 independent parts. To try it: go to antigravity.google.com, open a new conversation, add a project folder, select Gemini 4 Carbon, set permissions to auto, type '/swarm', paste the prompt, and open the swarm UI.

## Transcript

Google Anti-Gravity just got Gemini 4 Carbon and a brand new swarm mode, so I used it to build a hive. One prompt, nine AI agents. Each one builds a piece of a real business app, while two inspector agents check every single cell.
This is the hive. Every being is a real agent. Every cell is real code. And by the end, you'll have three working systems. Here's what happened. On November 10th, Google put Gemini 4 Carbon inside anti-gravity.
Carbon is their strongest model for long software jobs, and Google says it's number one on CursorBench 4.0. And with it came Slash Swarm. It's in preview in the anti-gravity app and the CLI.
Before this, one chat meant one agent doing one thing at a time. Now, your chat becomes the coordinator. You type Slash Swarm, then your task. The coordinator splits the job, spawns parallel workers, and each worker gets its own isolated copy of the project.
Then come the verifiers. Their whole job is to challenge the work. They run the tests, open the app, and send anything broken back. Everyone talks on a shared board. And when an agent has nothing to do, it sleeps, so it isn't burning anything.
On the right, you get the swarm pane. Board is the live feed. Members shows every agent, and you can jump in and steer one or pin it to a model. And graph lets you scrub back through the whole run.
So, let's build. A member in the AI Profit boardroom asked this week, I've sold a lot of websites. How do I build my own client system instead of renting one? So, comb one is the client comb, a client system for a web design agency.
Here's the prompt. Slash swarm build client comb Four workers one module each leads follow bookings and reviews A shared shell joins them together and the verifiers must open it in a real browser and reject anything that breaks Watch the hive
The coordinator briefs nine agents. The shell worker lays down the design system first. Then the four module cells start filling with honey, and that's real code landing on disk. Lint hits 22 kilobytes.
Follow-ups, bookings, and reviews all climb at the same time, in parallel. Then the inspectors fly in. Verifier 1 checks every button, the console, and reloads. Verifier 2 measures every element at 1280x720 for overlaps.
Both passed. Zero console errors, zero overlaps. The cells get sealed. 24 minutes, 11 files, and here's what it built. A pipeline you can drag leads through. Follow-ups due today with the message already drafted.
A booking calendar. And a review request funnel. Quick one. If you want the exact swarm prompts from this video, plus the hive itself and the setup guide, they're inside the AI Profit Boardroom.
You also get the faceless video skill I use to make videos like this, my best anti-gravity and Gemini setups, and a community of thousands of business owners building with AI every day. And you can post your questions and get help.
The link is in the comments and the description. Or go to AIProfitBoardroom.com. Cone 2. Another member is running their first AI audits this month, on-site with two client businesses.
So the audit comb is an AI audit kit for an automation consultant. Same move. Slash swarm. Four workers. Intake, process map, scorecard, and report. The intake worker builds a guided interview.
The team the tools and the hours each task takes every week The process map worker draws the client workflows as connected steps and flags every manual handoff The scorecard worker plots every automation idea on impact versus effort
and picks the top three by itself. And the report worker pulls all of it into a client-ready audit with a 30-, 60-, 90-day roadmap and a print-to-PDF button. The hive fills up again.
477 kilobytes of code in about 32 minutes. And again, the inspectors pass every cell before it gets sealed. For the example agency in the demo, it finds 48 and a half hours a week that could be automated.
And that's the stream you walk your client through on site. Comb3 is for coaches and community owners. One of the busiest threads in the boardroom this week was simply, how does anyone keep up? So the Keeping Up Comb is a weekly update desk
for a coach who runs an online community. And I seeded it with only the real anti-gravity updates from this week. Nothing made up. Four workers again. Inbox collects the updates. Triage sorts each one into Use Now, Try Later, or Ignore for a coaching business.
Actions turns the Use Now pile into this week's top three moves. And Member Post writes a short community post about them with a phone preview. And this is the run where the inspectors really prove themselves.
Verifier 1 passed 54 checks, then sent the shell worker back to fix a missing file. The cell lights up red. Verifier 2 found two visual issues in the member post screen and sent that worker back too.
Both workers fixed their cells. Both verifiers checked again. 57 out of 57. Zero errors. Sealed. That's the whole point of the verifiers. Nobody had to tell them to look.
They caught it sent it back and checked again all by themselves Now a few things I learned running these First be specific about who owns what One module per worker That what lets them build at the same time without stepping on each other
Second, write the verify rules into the prompt. Real browser, no console errors, no overlaps, data survives a reload. That's what made the checks real. Third, Swarm is quota heavy.
Every worker is its own agent. so save it for jobs that really split into parts. And fourth, use the graph tab. After a run, you can scrub back through the whole thing and see who sent what to who and when.
So, what would you swarm next? If you run a SaaS, try an onboarding flow. Welcome screens, a setup checklist, and help docs, one worker each. If you sell a course, a student portal with lessons, progress, and a question box.
And if you run a store, a product page, a size guide, and an FAQ, FAQ built side by side. The rule is simple. If the job splits into three or four parts that don't depend on each other, it's a swarm job. Here's how to try it. Go to antigravity.google.com. It
runs on Mac, Windows, and Linux. Open it, click New Conversation, and add your project folder. In the model picker, pick Gemini 4 Carbon. Carbon is rolling out to Google AI Ultra first.
Set permissions to auto so it doesn't stop to ask on every file. Then type slash swarm. Paste your prompt and hit enter. When the coordinator replies, click open the swarm UI.
Watch the board and check the members tab to see who's working. If you want all three swarm prompts, the hive and the step-by-step guide, they're inside the AI Profit boardroom, along with my faceless video skill and the latest anti-gravity setups.
The link is in the comments and the description. or go to AIprofitboardroom.com. See you in the next one.
