Antigravity

Confidence 0.75 · 5 sources · last confirmed 2026-10-01

Google’s single agent harness, built by the Windsurf team that joined Google. Per Logan Kilpatrick in the Sequoia interview, it is “a lot of things” — a core IDE, an agent-first web surface, a CLI, an SDK, and a managed agent available through the Gemini API — designed to “meet developers wherever they are.”

Why it matters to the wiki

• The new through line for Google. The same Antigravity harness powers search, the Gemini app, Google Cloud, and AI Studio — taking over the connective-tissue role that the Gemini model/API used to play across Google’s ~50 products. A concrete instance of the agent-harness concept at hyperscaler scale.
• Coding harness → general-purpose harness. Kilpatrick: “coding has proved to be the general-purpose agent harness.” ~80% shared base, specialized per use case (vibe coding vs. a 24/7 consumer agent).
• Named as a harness, by Google, in Google’s own harness explainer. In [[2026-07-16-baugues-thurium-google-cloud-what-is-an-agentic-harness|What is an Agentic Harness?]] (July 2026), Greg Baugues lists it among “the most popular agentic harnesses today — so like a Claude Code or an Antigravity or a Codex” — and the point he makes with the list is the harness/interface decoupling: each of the three carries several interfaces over one runtime, so the product name is not the unit of analysis. Note the ordering: Google’s own DevRel names a competitor’s harness first.
• In an evaluation role, not a coding one. In [[2026-09-30-zamora-feroz-why-your-ai-agent-fails-in-production|Google Cloud Tech’s AI Agent Clinic (September 2026)]], Antigravity reads a LangGraph agent’s source and traces, draws its architecture as a Mermaid diagram, runs the agent on three more repositories to generate test scenarios, writes the trace converter that lets Google’s ADK-built eval toolkit read the agent, and drafts the eval metrics. The developer reviews the metrics; no check of the judges against human labels is shown.
• Already in the corpus. Named (“Anti-gravity”) alongside Claude Code, Cursor, and OpenCode in 2026-05-04-rethinking-agents-harness-is-all-you-need as one of the harnesses that drives same-model performance variance.

Built within Google DeepMind; the model-eats-the-harness tension (Kilpatrick’s own thesis) applies to Antigravity as much as to any harness.