Antigravity logo

[ AI tools · Development ]

Antigravity Agents That Prove Their Work

Google’s Antigravity runs several coding agents at once across the editor, terminal and browser, and each one shows screenshots, recordings or test reports of what it did. We use it most on frontend and test work.

Agent Manager

3 agents · 1 reviewer

agent 1

Build the pricing page from the design

  1. Plan
  2. Edit
  3. Open in browser
  4. Compare to design

Artifact · Screenshot

agent 2

Fix the mobile menu overlap on iPhone sizes

  1. Reproduce
  2. Fix CSS
  3. Test 3 sizes
  4. Record

Artifact · Browser recording

agent 3

Add end-to-end tests for checkout

  1. Plan
  2. Write tests
  3. Run
  4. Report

Artifact · Test report

  • Adds item to cart
  • Applies discount
  • Pays with card
  • Shows receipt

Example agents. Each one shows its work before a person approves it.

Google’s Agent-First Development Platform

Antigravity grew from an agent-first IDE into a platform for orchestrating agents alongside your editor. Its strength is verification: agents check their own changes in a real browser.

Agent Manager
  • Checkout page from FigmaWorking
  • Fix header on small phonesReview
  • E2E tests for sign-upDone

1Agent ManagerA mission-control view to start, watch and steer several agents working on separate tasks.

Editor
export function Header() {
  return (
    <nav className="flex gap-4">
      <Logo />
      <Menu />
    </nav>
  );
}

2EditorA full code editor for the parts you want to write yourself, with agents a keystroke away.

Browser control
localhost:3000/checkout Pay button visible at 390px

3Browser controlAgents open your app in Chrome, click through it and check the result like a tester would.

  • 04ArtifactsPlans, task lists, screenshots and recordings that show what an agent did, so you review evidence, not logs.
  • 05Antigravity CLIThe terminal client that replaced Gemini CLI for individual developers in June 2026.
  • 06Gemini and other modelsBuilt around Gemini models, with other providers' models available for some tasks.
[ Artifacts ]

Six Kinds of Evidence an Agent Hands Back

Reviewing an agent’s work is faster when it shows you the result instead of a wall of logs. These come with every task.

  • 01ArtifactImplementation planWhat the agent intends to change and why, approved before it starts.
  • 02ArtifactTask listThe steps it will take, ticked off as it goes, so you can see progress at a glance.
  • 03ArtifactScreenshotsThe page as built, side by side with the design, at the sizes you care about.
  • 04ArtifactBrowser recordingA replay of the agent clicking through the flow, including the failure it fixed.
  • 05ArtifactTest reportWhich end-to-end tests ran, what passed and what changed.
  • 06ArtifactWalkthroughA short summary of the finished change for the reviewer, with links to each file.
[ Leaving Gemini CLI ]

Moving From Gemini CLI to Antigravity CLI

Google retired Gemini CLI for individual and free users on 18 June 2026 in favor of Antigravity CLI. Teams with scripts built on it need a careful switch; here is ours.

Web developer's desk with a monitor and laptop showing websites in progress
Agent artifact01 / 04Screenshot diff: 3 breakpoints matchRecording: before and after the fixCheckout journey: 12 browser tests passFirebase rules updated and deployed to staging
[ How we use it ]

Where Antigravity Earns Its Place on Our Projects

  1. 01

    Frontend builds from designs

    Agents build pages and compare screenshots to the design at several screen sizes before anyone reviews them.

  2. 02

    Visual and responsive fixes

    Bugs reproduced in a real browser, fixed and recorded, so the reviewer sees the before and after.

  3. 03

    End-to-end test coverage

    Agents write and run browser tests for key journeys and report what changed.

  4. 04

    Google Cloud projects

    Work that touches Firebase or Google Cloud, where Gemini and Google tooling fit naturally.

[ Controls ]

How We Keep Several Agents in Check

  • Review before action

    Agents ask before running terminal commands or changing files outside the task, unless we allow it.

  • Separate workspaces

    Each agent works on its own branch or worktree, so parallel tasks do not collide.

  • Evidence on every task

    No change is accepted without the artifact that shows it works.

  • Human merge

    A senior engineer reviews and merges, with the agent's plan and evidence attached.

Pick Your Starting Point

Packages from $1,500; custom projects from $8,000.

Add an engineer

Dedicated Developer Trial

A senior engineer for a month who runs Antigravity agents on your frontend and test backlog.

$3,000 to $5,000/mo1 month

See the details

Antigravity, Asked

Ask Your Own
  1. What is Google Antigravity?

    Antigravity is Google's agent-first development platform. It lets you run and supervise several coding agents across an editor, terminal and browser, with artifacts such as screenshots and recordings that show what each agent did.

  2. Is Gemini CLI still available?

    Google shut down Gemini CLI for individual and free users on 18 June 2026 and points them to Antigravity CLI. Enterprise customers were exempt from the shutdown. The open-source repository still exists.

  3. Which models does Antigravity use?

    It is built around Google's Gemini models and also offers some models from other providers. We pick per task and test on your codebase.

  4. What is Antigravity best at?

    Frontend and test work, where an agent can open your app in a browser, check the result against the design and record what it did. That evidence makes review much faster.

  5. Is it safe to let agents control a browser?

    We run them against local or staging environments, never production, with test accounts and no access to real customer data. Every change is reviewed and merged by a senior engineer.

  6. Can you help us move off Gemini CLI?

    Yes. We inventory your scripts and CI jobs, move context files and MCP servers, run both tools side by side and switch over job by job.

Antigravity runs on Gemini. To build agents into your own product, use Google ADK; for large refactors from the terminal, see Claude Code; and for test automation, our testing and QA work.

[ Contact ]

Let's build your next product together

Book a free strategy call and leave with a clear plan and estimate. No commitment.

  1. 01Pick a time that suits you
  2. 0230 minutes on scope, stack, timeline and budget
  3. 03Fixed-price proposal, NDA on request
Prefer email? contact@sajaltech.com