Antigravity is Google's agent-first development platform, launched in public preview on 18 November 2025 alongside Gemini 3 Pro. It was built by the team Google brought in through a $2.4B licensing deal with Codeium/Windsurf in July 2025, which moved CEO Varun Mohan, co-founder Douglas Chen and key engineers into Google DeepMind.
The "agent-first" framing is substantive rather than marketing. Antigravity has two surfaces: a conventional Editor view with an agent sidebar, and an Agent Manager view that behaves like mission control for several agents working in parallel, each optionally isolated in its own git worktree. Agents report through Artifacts — implementation plans, task lists, screenshots and recordings of real browser sessions — so review happens against deliverables instead of raw tool-call transcripts. Version 2.0, shipped at I/O in May 2026, added a Go CLI, an SDK for custom agents, scheduled background tasks and voice commands; Google retired Gemini CLI for consumers the following month and folded it into Antigravity.
Two things temper the picture. First, security researchers demonstrated indirect prompt injection within a week of launch — instructions hidden in tiny text on a web page caused agents to read .env files and exfiltrate secrets — and Google reportedly treated some of this as expected behavior; we found no public advisory closing it out. Second, the March 2026 shift to AI credits triggered sustained complaints on Google's own developer forum, with developers reporting their usable weekly token budgets collapsing by an order of magnitude. Google does not document what a credit buys, which makes cost planning guesswork.
Key Benefits
- Parallel agents with real isolation: Worktree mode lets several agents work simultaneously without stepping on each other's changes.
- Verifiable output: Artifacts and recorded browser sessions give a concrete basis for reviewing what an agent actually did.
- Genuine model choice: Claude Opus 4.6 and Sonnet 4.6 sit alongside Gemini on free and Pro tiers — a notable openness for a first-party vendor tool.
- Ecosystem depth: CLI, SDK and integrations with Google AI Studio, Firebase and Android make it a plausible hub for teams already on Google infrastructure.
Use Cases
- Parallel feature development — Assign several agents to independent tasks across isolated worktrees and review their plans and diffs from the Agent Manager.
- Front-end work with visual verification — Agents drive Chrome, exercise the UI and attach screenshots and recordings, so visual regressions surface before review.
- Scheduled maintenance tasks — Background agents handle dependency updates, test triage and routine refactors on a recurring schedule.
- Custom agent development — The SDK lets teams build purpose-built agents on Google's harness rather than assembling orchestration from scratch.
Security note: given the documented prompt-injection findings, avoid pointing Antigravity agents at untrusted web content in repositories that contain credentials.
