Google Antigravity

Learning Objectives

• Understand how Antigravity's agent-first architecture differs from AI-assisted editors
• Identify the Manager View's capabilities for dispatching and monitoring multiple agents
• Evaluate when Antigravity is the right choice versus Cursor, Claude Code, or GitHub Copilot

What Is Google Antigravity?

Google Antigravity is an agent-first IDE from Google DeepMind, announced in November 2025 alongside Gemini 3 and substantially expanded as Antigravity 2.0 at Google I/O 2026. Built as a VS Code fork, Antigravity takes a fundamentally different approach to AI-assisted development: instead of a single AI assistant, it lets you dispatch 5 or more agents simultaneously through a Manager View dashboard.

The Manager View is Antigravity's defining feature. You assign tasks to individual agents — "write the API routes," "create the test suite," "set up the database schema" — and they work in parallel. Each agent produces Artifacts: task lists, implementation plans, screenshots, and even browser recordings that document what it did and why. You supervise from the Manager View, reviewing artifacts, redirecting agents, and merging their work.

Antigravity 2.0 adds Gemini 3.5 Flash and Gemini 3.5 Pro as the default routing pair, with parallel subagents that compress multi-week workflows into hours — and a unified backend shared with a new Go-based Antigravity CLI, the terminal-side companion that absorbs the Gemini CLI installed base ahead of Gemini CLI's June 18, 2026 sunset.

Antigravity supports multi-model selection: Gemini 3.5 Pro, Gemini 3.5 Flash, Claude Opus 4.7, Claude Sonnet 4.6, and GPT-OSS-120 billion. Different agents can use different models — use a fast model for simple tasks and a powerful model for complex reasoning.

Currently in free public preview for individual developers.

Pricing

Antigravity is free during the public preview. Google has not yet announced pricing for the general availability release, but enterprise and team plans are expected.

Core Capabilities

Manager View

The Manager View is a dashboard for dispatching, monitoring, and coordinating multiple AI agents. Assign tasks in natural language, see each agent's progress in real time, review their artifacts, and merge their outputs. Think of it as a project manager's view into an AI development team.

Multi-Agent Dispatch

Launch 5 or more agents simultaneously, each working on a different part of your project. Agents operate independently — one can write frontend components while another builds API routes and a third generates tests. This parallel approach dramatically reduces the time for complex, multi-faceted development tasks.

Artifacts

Every agent produces artifacts — structured outputs that go beyond code. Task lists show what the agent planned. Implementation plans document its approach. Screenshots capture the visual state of UI work. Browser recordings show how the agent tested interactive features. Artifacts make agent work transparent and reviewable.

Multi-Model Selection

Choose which model each agent uses. Assign Gemini 3.5 Pro for complex architectural decisions, Gemini 3.5 Flash for quick edits, Claude Opus 4.7 for deep reasoning, or GPT-OSS-120 billion for open-source tasks. Model selection is per-agent, not global.

VS Code Foundation

Built on the VS Code codebase, Antigravity supports existing extensions, themes, and keybindings. Developers familiar with VS Code can start immediately — the difference is the agent infrastructure layered on top.

Antigravity CLI — Terminal Companion

Antigravity CLI is the Go-based terminal companion that ships alongside the desktop IDE in Antigravity 2.0. It exposes the same Agent Skills, Hooks, Subagents, and Extensions (renamed plugins) that developers configure inside Gemini CLI today, plus asynchronous workflows that let multi-step tasks run in the background while the terminal stays responsive. A shared unified backend means a session started in the CLI can be picked up in the desktop Manager View — useful for handing off a long-running agent task from a CI runner or remote shell to a local IDE.

The CLI rewrite is also Google's explicit migration path for Gemini CLI, which sunsets June 18, 2026 for Google AI Pro, Ultra, and free-tier users. Existing Gemini CLI configurations port forward directly to Antigravity CLI — see the Gemini CLI lesson for the full migration sequence.

Multi-Week Workflows in Hours

Antigravity 2.0 emphasizes multi-week workflows compressed into hours through parallel subagent spawning. A single top-level agent decomposes a large engineering task — a payments integration, a multi-service migration, a security audit — into independent subtasks that subagents execute concurrently. The Manager View shows progress across all parallel branches; artifacts roll up to a single review surface. Google's stated benchmark for the 2.0 release is that production teams running this pattern have completed work in days that previously took weeks — a claim that's still maturing in independent benchmarks but consistent with the platform's parallel-by-default design.

Strengths

• Multi-agent parallelism: Dispatch multiple agents working simultaneously — unique capability among AI IDEs
• Manager View: Centralized dashboard for overseeing, redirecting, and merging agent work
• Artifacts: Transparent documentation of what agents did, how, and why — critical for review
• Multi-model choice: Mix models across agents based on task requirements
• Free during preview: Full access at no cost while in public preview
• Google ecosystem: Native integration with GCP, Firebase, and Google Cloud services

Limitations & Considerations

• Public preview: Not yet generally available — features may change and stability may vary
• No pricing clarity: Enterprise and team plan costs are unknown until GA
• Newer platform: Less mature than Cursor, Copilot, or Claude Code — smaller community and fewer resources
• Agent coordination complexity: Managing multiple agents effectively requires a new mental model for development
• Google account required: Tied to Google authentication and ecosystem

Best Use Cases

When to choose alternatives:

• Mature AI IDE today → Cursor (established, 360K+ paying customers, multi-model)
• Terminal-native workflows → Claude Code (file system access, MCP, git)
• GitHub-native development → GitHub Copilot (Issues, PRs, coding agent)
• Budget-conscious AI IDE → Windsurf ($15/month Pro with full features)

Getting Started

1. Visit developers.google.com/antigravity and sign up for the public preview
2. Download and install Antigravity — import your VS Code settings and extensions
3. Open a project and explore the Manager View from the sidebar
4. Dispatch your first agent: describe a task and assign a model
5. Try multi-agent dispatch: assign two or more tasks simultaneously and monitor from the Manager View
6. Review artifacts produced by each agent before merging their work

Key Takeaways

• Google Antigravity is an agent-first IDE that lets you dispatch and manage multiple AI coding agents simultaneously — a fundamentally different approach from single-assistant editors
• Antigravity 2.0 (May 2026) adds parallel subagents that compress multi-week workflows into hours, a unified backend with a new Go-based Antigravity CLI, and default routing across Gemini 3.5 Pro and Gemini 3.5 Flash
• Antigravity CLI is also the supported migration path for Gemini CLI, which sunsets June 18, 2026 for AI Pro, Ultra, and free-tier users
• The Manager View, artifacts, and multi-model selection give developers oversight and control over parallel agent work
• Currently free in public preview; pricing for GA and enterprise plans is not yet announced
• Best suited for complex, multi-faceted development where parallelism accelerates delivery; for a mature AI IDE today, consider Cursor