AI Summary
Google Antigravity's version 2.32.0 update, released on November 13, 2026, introduces two major features: Swarm, which launches a team of specialized AI agents that collaborate on a live message board, and Dynamic Workflows, which generates a reviewable Python script to orchestrate sub-agents. The video demonstrates how these tools can automate complex tasks like building and testing a landing page, and provides guidance on when to use each feature.
Chapters
Google Antigravity's Swarm update allows a single prompt to launch a team of AI agents that plan, build, test, and verify work in a browser.
Antigravity version 2.32.0 was released on November 13, 2026, featuring Dynamic Workflows and an expanded Swarm command.
Swarm runs a team of specialized agents simultaneously, like a mini company. Agents plan, build, test, and review, communicating on a live message board. Different models can be assigned to different roles.
Real work involves multiple steps (plan, build, test, review). Swarm handles the loop automatically, so you give the goal and the team figures out the roles.
A single prompt can instruct Swarm to design a landing page, write tests, run them, review changes, verify in a browser, and send a summary. Agents work in parallel on different files.
The 'run workflow' command makes the agent write a short Python script that acts as the plan. Sub-agents run in parallel or sequentially with clean handoffs. You can review, edit, and save the script for reuse.
Workflow scripts prevent AI drift by defining steps and order, giving you control while maintaining speed. The command-line version bundles Python, so no separate installation is needed.
Sub agents handle delegated tasks in a clean space. The browser sub agent can open pages, click through flows, check console/network errors, take screenshots, and verify fixes.
Google shared examples of Swarm building a 3.js Eiffel Tower exhibition and an office scene with ~96,000 triangles, showing it's built for long, complex tasks.
Swarm requires a paid plan. Dynamic Workflows is available on Google AI Pro and Google AI Ultra.
Use Swarm for open-ended jobs where agents need to collaborate and react. Use run workflow for clear, repeatable tasks. You can combine both: start with Swarm to build, then use workflow to check.
The update transforms AI from a single-task tool into a collaborative team, with Swarm for flexible, open-ended work and Dynamic Workflows for repeatable, controlled processes. This enables users to automate complex workflows and verify results without manual oversight.
Mentioned in this Video
Tutorial Checklist
💡 Key Takeaways
Swarm handles the loop
Shifts AI from single-task to autonomous team coordination, a key paradigm change.
01:43Workflow scripts prevent AI drift
Addresses a common pain point of AI unpredictability with a reviewable, repeatable plan.
04:27Swarm handles complex builds
Demonstrates scalability with examples like a 96,000-triangle scene, proving it's not just for small tasks.
06:25Swarm vs Workflow decision guide
Provides a clear, practical rule for choosing between the two features, directly actionable for users.
07:10Full Transcript
[00:02] Google Anti-gravity's Swarm update is next level. One prompt now launches a whole team of AI agents. They plan, they build, they test, they even check your site in a browser. It's wild. Stay with me because I'll show you which command
[00:17] most people will miss. Google Anti-gravity just dropped version 2.3 Anti-gravity just dropped version 2.3 2.0. It came out on November 13th, 2026. And it has two big things inside. Dynamic workflows and a bigger swarm
[00:32] command. Here's why that matters. Most people use AI one task at a time. You ask, it answers. You ask again. That's slow. This update changes it. One prompt
[00:44] can now turn into a full team of agents. I'll show you how to use it for the AI profit boardroom step by step. And at the end, I'll tell you which command to use and when. Don't skip that part. Let's start with Swarm. What is it?
[00:59] Swarm lets you run a team of specialized agents at the same time. Think of it like a mini company inside your computer. One agent plans, one builds, one tests, one reviews, and they all talk to each other. Where do they talk?
[01:14] On a live message board. You can watch it happen. Agents post updates. They react to each other. They fix problems together. Here's the part I love. You can give different models to different roles. So your planner can use one model
[01:28] and your builder can use another. Swarm works in the desktop app. And it works in the command line, too. So whether you love clean apps or love your terminal, you're covered. Now, why is this a big deal? Because real work is never one
[01:43] deal? Because real work is never one step. It's plan, build, test, review, and check again and again. Normally, you do all of that yourself or you babysit one agent through every step. With Swarm, the team handles the loop. You
[01:58] give the goal, they figure out the roles. That's the shift. Let me show you a real example. Say you want a new landing page for the AI profit boardroom. You type something like this. Swarm, design a new landing page for the
[02:12] AI profit boardroom that converts. Show the value of AI automation. Explain every benefit of our AI automation community so we get more customers, leads, and traffic. Write tests for the form. Run the test suite, review the
[02:27] changes, then verify the whole page in the browser, and send me a short summary when you're done. That's one prompt. Now, watch what happens. A planner agent breaks the job into steps. Two builder agents work on different files at the
[02:42] same time. A tester runs checks on the signup form. A reviewer looks at the changes and flags anything weird. And a browser agent opens the page, clicks around, and checks it looks right. All at once, all talking on the message
[02:57] board. That's what turns one prompt into a finished checked page. Now, quick pause. If you want to save time and automate your business with AI tools like Google Anti-Gravity, you need to see what's inside the AI Profit
[03:11] Boardroom. It's our paid AI automation community. Inside, we show you how to use tools like Swarm to build pages, systems, and workflows without doing everything by hand. You get step-by-step training, real examples, and a group of
[03:27] people who are doing this every day. If you want to stop wasting hours on busy work, the link is in the comments and description. Okay, back to the update. Now, let's talk about the second feature, dynamic workflows. This one is
[03:41] very different and honestly it's a little sneaky good. You use the command run workflow. When you do, the agent writes a short Python script. That script is the plan. It tells the sub agents what to do. Some run at the same
[03:56] time. Some run one after another and they pass clean handoffs to each other. Here's the key. You get to review that script before anything runs. You can read it, you can change it, and you can save it to run again later. So, it's not
[04:11] a mystery. It's a plan you can see. Why does that matter? Because AI can drift. You know that feeling? You give a long prompt and the AI wanders off. It does its own thing. A workflow script fixes that. The steps are written down, the
[04:27] order is clear, so you get control, and you still get the speed. And here's a nice bonus. The command line version uses Python that's bundled inside it. So you don't even need to have Python installed. That's a small thing, but it
[04:41] removes a big headache. Let me give you a practical example for the AI profit boardroom. Say your team just updated the website. Lots of pages changed. You type this, run workflow, review every changed file on the AI profit boardroom
[04:55] changed file on the AI profit boardroom site, run the tests, fix any failures, then give me a final change report. Now, anti-gravity writes the script. Maybe it has stages like analyze changes, review files, run tests, fix failures, and
[05:10] final report. You read it, you click approve, then the agents go to work. One reviews each file, another runs the tests, another fixes what broke, and you get a clean report at the end. That's a repeatable system. You can use the same
[05:26] workflow every time you update the site. Now, there are two more features I want you to know about. First, sub agents. This is the simple version. Your main agent hands off a task to a separate agent. Maybe research, maybe a build,
[05:40] maybe a test. That agent works in its own clean space at the same time as the others, so nothing gets muddy. Use this when you have one task to delegate and you don't need a big setup. Second, the browser sub agent. This one is so
[05:55] useful. The agent can open a page. It can click through a flow. It can look at console errors and network errors. It can take screenshots. And it can check that a fix actually worked. So for the AI profit boardroom, you could say,
[06:10] "Open our signup page, click through the whole flow, find anything broken, fix it, and show me a screenshot when it works." That's web testing without you sitting there clicking for an hour. Now, how good can Swarm get? Google shared
[06:25] some examples with the Gemini 4 carbon launch. In one, Swarm built a 3.js Eiffel Tower exhibition. In another, it rebuilt an office scene with about 96,000 triangles. Those are big, heavy builds.
[06:40] And that tells you something. This isn't just for tiny tasks. It's made for long, tricky jobs where many agents need to work together. Okay, now the boring but important part. Who can use this? Swarm needs a paid plan. Dynamic Workflows is
[06:57] available on Google AI Pro and Google AI Ultra. So, check your plan before you try it. I don't want you getting stuck and thinking it's broken. It's usually just the plan. So, here's the big question I promised you. Which one
[07:10] should you use? Easy. Use Swarm when the job is open-ended. When agents need to talk, argue, review each other, and react. Like building a new landing page for the AI profit boardroom from scratch. Use run workflow when the job
[07:25] is clear and you want it done the same way every time. Like checking every changed file before you publish. Swarm is the live team. Run workflow is the saved playbook. And the best part you can use both. Start with swarm to build
[07:40] something. Then use run workflow to check it the same way every time. Let me quickly recap so it sticks. One. Anti-gravity 2.32.0 Anti-gravity 2.32.0 came out on November 13th, 2026. Two,
[07:55] Swarm gives you a live team of agents on a shared message board. Three, run workflow writes a Python script you can review, save, and reuse. Four, [snorts] sub agents and the browser agent handle delegated tasks and real web checks. And
[08:11] five, use Swarm for open-ended work and run workflow for repeatable work. That's the whole update. Now, if you liked this, here's how to go deeper. If you want to learn how to save time and automate your business with AI tools
[08:26] like Google Anti-Gravity, join the AI Profit Boardroom. It's where we break down the newest AI updates and turn them into simple systems you can use. You get training, templates, and real examples, all built around AI automation. The link
[08:41] is in the comments and description. And if you want the full process, the SOPs, and 100 plus AI use cases like this one, join the AI Success Lab. It's our free community with new AI training released daily. Inside, you'll get the video
[08:56] notes from videos like this one, plus access to our community of over 90,000 members who are all using AI to grow their businesses. The link is in the comments and the description if you want to check it out. I'll see you in the
[09:10] to check it out. I'll see you in the next one.