Google Antigravity launched on November 18, 2025 alongside the Gemini 3 Pro model, giving developers a new AI IDE that takes an “agent-first” approach to coding. Rather than just sprinkling autocomplete and chat into an editor, Google is trying to redesign the development environment around autonomous agents that can plan, execute, and verify end-to-end tasks across your editor, terminal, and browser.
This is NEWS CONTENT, not a longform tutorial. Below is a concise, developer-focused first look at Google Antigravity, how it uses Gemini 3, and how it compares to tools like GitHub Copilot as of November 2025.
What is Google Antigravity?
Google Antigravity is a new agentic development platform and AI IDE released in public preview for Mac, Windows, and Linux on November 18, 2025. At its core it looks and feels like a familiar code editor, but it is built around “agent-first” workflows: you spin up agents that can autonomously write code, run commands in a terminal, control a browser, and report back with verifiable “artifacts” like plans, test results, and recordings.
Antigravity is powered primarily by Gemini 3 Pro (the new flagship model with a 1M-token context window and strong coding/agentic benchmarks), but it also supports Anthropic Claude Sonnet 4.5 and OpenAI’s GPT‑OSS in the preview, giving developers model choice inside a single environment.
Key features and architecture
Google describes four core tenets of Antigravity’s design: trust, autonomy, feedback, and self-improvement. These are expressed through several concrete features:
- AI IDE core: VS Code–style editor with tab autocompletion, inline natural-language commands, and a context-aware agent sidebar.
- Higher-level, task-based abstractions: Work is grouped into tasks, each with a visible status, plan, and associated artifacts instead of a stream of opaque tool calls.
- Cross-surface agents: Agents can control the editor, terminal, and browser in sync (using Gemini 3 plus the Gemini 2.5 computer-use model) to implement and validate features end-to-end.
- Agent Manager (“mission control”): A separate manager view for spawning, orchestrating, and supervising multiple agents and workspaces asynchronously.
- Feedback and learning: You can comment on artifacts (text and screenshots) Google‑Docs style; agents feed this plus their own history into an internal knowledge base.
- Model flexibility: Gemini 3 Pro-preview is the default, but you can also choose Claude Sonnet 4.5 or GPT‑OSS for specific agents.
How “agent-first” development works in practice
Most AI IDEs today embed an assistant inside a single surface (e.g., inline Copilot suggestions in VS Code). Antigravity flips this by making the agent the primary object and treating the editor, terminal, and browser as tools it can control.
- You describe a task in natural language (for example, “Add a real-time flight tracker page with map and tests”).
- The agent creates a task with a plan (steps, files to touch, tests to run) and shows this plan as an artifact.
- It then codes in the editor, launches your app in the terminal, and drives the browser to test behavior, using Gemini 3’s tool-use and reasoning.
- Throughout, it generates artifacts (implementation plans, screenshots, browser recordings, walkthroughs) that you can inspect and comment on.
- Once the task is complete, you can accept, edit, or ask it to iterate – with your feedback feeding into its knowledge base.
This design aims to support longer-running, partially autonomous workflows that you can drop in and out of, rather than constant synchronous back-and-forth prompting.
How it compares to GitHub Copilot and other AI IDEs
As of November 2025, GitHub Copilot remains deeply integrated into VS Code and JetBrains, and GitHub has been rolling out more “agentic” features (Copilot Agents, Copilot CLI, code review, etc.). But the core experience is still editor-first. Antigravity instead ships as a dedicated, AI‑first IDE whose entire UX is centered on agents and tasks.
| Aspect | Google Antigravity (2025 preview) | GitHub Copilot (2025) |
|---|---|---|
| Primary model | Gemini 3 Pro-preview (1M context, Jan 2025 cutoff) plus Claude Sonnet 4.5, GPT‑OSS options | GitHub-hosted models & OpenAI family (latest Copilot stack, evolving through 2025) |
| Form factor | Standalone, AI-first IDE with dual views: Editor and Agent Manager | Extensions and chat inside existing IDEs (VS Code, JetBrains, Neovim, etc.) |
| Agent orchestration | Built-in multi-agent “mission control” across workspaces | Agent sessions and CLI; orchestration mainly via GitHub/Copilot surfaces |
| Cross-surface control | First-party control of editor, terminal, and browser with task-level artifacts | Strong editor integration; terminal/browser control emerging but not IDE-native |
| Transparency | Task-based artifacts (plans, screenshots, recordings, walkthroughs) | Chat history, diffs, inline suggestions; less emphasis on task artifacts |
| Pricing (launch) | Free public preview with “generous” Gemini 3 Pro limits | Paid subscription per user; some enterprise SKUs and trials |
If you already live in GitHub’s ecosystem and prefer to stay inside VS Code, Copilot will feel more natural. If you’re looking for a dedicated environment to experiment with multi-agent, cross-surface workflows, Antigravity is designed specifically for that.
Is Google Antigravity right for your workflow?
Because Antigravity is in free public preview, it’s low-friction to test. It’s most compelling if:
- You want to explore agent-first development: spinning up agents to run longer-lived tasks while you context-switch.
- Your work spans editor + terminal + browser, and you’d benefit from an agent that can autonomously drive all three.
- You care about verifiable AI workflows, with plans, test runs, and browser recordings you can audit rather than just blind diffs.
- You want to try Gemini 3 Pro-preview in an environment tuned for its reasoning and tool-use strengths.
On the other hand, you may want to wait if you need tight integration with existing corporate tooling, deeply customized VS Code setups, or mature enterprise controls that have been battle-tested over years; Antigravity is explicitly labeled an experiment and will evolve quickly.
What to watch next
As of November 2025, Gemini 3 Pro-preview is the only Gemini 3 model exposed to developers, and Antigravity is positioned as the flagship playground for its “vibe coding” and agentic capabilities. Google has not yet committed to GA timelines, but the preview already ships on all major desktop platforms and is tightly coupled with the Gemini API, AI Studio, and Vertex AI.
For developers, the key questions over the next few months will be:
- How stable and predictable are long-running agents on real-world codebases?
- Does the task-and-artifact UX actually improve trust compared with traditional AI assistants?
- How quickly do integrations, plugins, and enterprise controls catch up to VS Code + Copilot and other mature tools?
Right now, Google Antigravity is best seen as a serious, free testbed for agent-first development powered by Gemini 3. If you’re evaluating AI IDEs for 2026 and beyond, it deserves a hands-on trial alongside GitHub Copilot, Cursor, and other agentic environments.
