Google Antigravity
Google Antigravity
One-line summary: Google's November 2025 "agent-first" development platform — announced alongside Gemini 3, with a dual Editor/Manager view for orchestrating parallel agents across editor, terminal, and browser.
What it is
An agentic development platform by Google, announced November 18–20, 2025 alongside Gemini 3. Per Google's own announcement (2026-04-21-autoresearch-best-ai-coding-tools), it aims to evolve the IDE toward an "agent-first future" with browser control, asynchronous interaction patterns, and agent-first product design.
Why it matters to this thread
Antigravity is Google's most serious play in the coding-agent market and represents a distinct architectural bet: where claude-code is terminal-first and cursor is IDE-first, Antigravity's central object is the Manager view for orchestrating multiple agents in parallel. That framing is closest to the fleet-dispatcher model in parallel-claude-workflow but made first-class in the tool itself.
Key facts (from 2026-04-21-autoresearch-best-ai-coding-tools)
- Announcement date: November 18–20, 2025.
- Default model: Gemini 3 Pro (with generous rate limits).
- Model bring-your-own support: Claude Sonnet 4.5, OpenAI GPT-OSS.
- Pricing: Free in public preview for individuals. Post-preview pricing not announced.
- Two surfaces:
- Editor view — IDE-like interface with agent sidebar for synchronous work.
- Manager view — dedicated control center for spawning and orchestrating multiple agents asynchronously.
- Distinctive feature: Artifacts — agents generate "tangible deliverables (task lists, screenshots, recordings) rather than raw logs."
- Agent knowledge base: Agents save useful context for improving future task performance.
- March 2026 update: AgentKit 2.0 — 16 specialized agents, 40+ domain-specific skills, 11 pre-configured commands covering frontend, backend, testing, debugging, SEO, and database management.
- Benchmark: SWE-bench Verified: 76.2% (MorphLLM).
- Adoption (JetBrains April 2026): 6% adoption — notable given it launched only ~5 months before the survey.
Strengths
- Multi-model support (Gemini 3, Claude 4.5, GPT-OSS) — not vendor-locked despite being a Google product.
- Artifact outputs (screenshots, recordings) materially easier to review than raw logs.
- Manager view is a novel primitive for agent fleets.
- Free public preview removes the pricing variable entirely during evaluation.
Weaknesses / concerns
- 6% adoption in 5 months is real but modest; not a runaway.
- Post-preview pricing is unknown, making commitment risky.
- No independent benchmark/RCT data yet — claims are primarily from Google's own announcement and vendor blogs.