← back to the archiveCover illustration for “Google Antigravity: an agent-first coding environment”
TERMday 173·2d ago·Published ·by Andy Padia

Google Antigravity: an agent-first coding environment

In short: Google Antigravity is an agent-first development platform from Google where AI agents plan, write and verify code and report back with evidence.

Google Antigravity is Google's agent-first development platform, announced on 18 November 2025. It pairs an editor with a Manager surface for running several AI agents that plan, write, run and test code across the editor, terminal and a browser. Agents report back with Artifacts such as task lists, plans, screenshots and browser recordings, so you can check their work.

Google Antigravity in plain words

Picture an IDE where you are less the typist and more the manager, assigning tasks to several AI workers and reviewing what they hand back.

How it works

You describe a task. An agent proposes a plan, then works across files, the terminal and a built-in browser to implement and verify it.

Google's developer blog describes the Manager surface as a dedicated interface to spawn, orchestrate and observe multiple agents working asynchronously across workspaces. Agents produce Artifacts so that verification does not depend on reading every log line.

At launch it was a public preview at no cost for individuals. By November 2026 the product site lists an app for parallel local agents, a CLI, an IDE and an SDK.

Example

As an illustration, "add a dark mode toggle to the settings page and verify it in the browser": the agent edits the component and styles, starts the dev server, opens the page in its browser, toggles the setting and attaches screenshots of both themes for your review.

Where you'll meet it

  • Developers trying agent-first workflows from Google
  • Teams already standardised on Gemini models and Google Cloud
  • Comparisons with Cursor, Claude Code and OpenCode
  • Front-end work where browser verification matters

What it changes for you

My view: the idea worth borrowing is verification by Artifact, where an agent shows evidence for its claim of success. For developers, the skill becomes writing tasks with clear acceptance criteria. For teams, running several agents in parallel raises review load, so decide who reviews what before scaling up.

Google Antigravity vs Claude Code

Antigravity began as an IDE-style environment centred on managing agents visually and has since added a CLI. Claude Code is Anthropic's coding agent, which began in the terminal. Try both on the same task before choosing.

Common mistake

My view: running many agents in parallel because the interface allows it. The real limit is how much you can review. Two well-specified tasks beat six vague ones.

Claude Code · OpenCode · Agentic coding · Multi-agent system · Browser agent

Run only as many agents in parallel as you can review properly.

#field-guide#coding-agents#developer-tools#google#ides

Sources

  1. antigravity.google/blog/introducing-google-antigravity
  2. developers.googleblog.com/en/build-with-google-antigravity-our-new-agentic-development-platform
  3. antigravity.google
← older drop
Speculative decoding: a draft model speeding up a big one
newer drop →
GraphRAG: retrieval over a knowledge graph of your data

related drops

explore all 557 drops →
← back to the archiveday 175