What is Google Antigravity?

Google's agent harness. The engine room underneath Gemini Spark, and Google's answer to the open harness layer that OpenCLAW and Hermes occupy.

If you have read what is a harness, you already know the formula: the model is the engine, the harness is the car. A model on its own is text in, text out. The harness is what gives it memory, tools, a plan, and the ability to run for hours without you babysitting it. Google Antigravity is Google's car. It is the thing that turns Gemini from a chatbot you talk to into an agent that does multi-step work on its own.

What it is

Google Antigravity is Google's agentic platform: the layer that wraps the Gemini model with planning, tool use, parallel sub-agents, and scheduling so it can complete real tasks instead of just answering questions. It launched in late 2025 as an agent-first coding environment. At Google I/O 2026 it grew into Antigravity 2.0, a standalone platform with a desktop app, a command-line tool, an SDK, a "Managed Agents" tier inside the Gemini API, and an enterprise deployment path.

Its signature trick is dynamic sub-agents. One agent can spin up a fleet of specialized agents that work in parallel · one planning, one writing, one testing, one debugging · and each of those can spawn its own helpers. The headline demo at I/O: Antigravity ran 93 parallel sub-agents to build a working operating system in roughly 12 hours for under $1,000. The same machinery runs tasks on Google's cloud in the background, on a schedule, with nobody watching.

That is the whole idea. Antigravity is the harness; Gemini 3.5 is the model it drives. Model plus harness equals agent.

Why CEOs care

You will rarely open Antigravity yourself. You care because it is the harness underneath Gemini Spark, the 24/7 agent Google is putting in front of millions of Workspace users. When Spark drafts your investor update while your laptop is closed, Antigravity is the part that knew which steps to run, in what order, and when to stop and ask you.

The bigger point is what it signals. Antigravity is Google planting its flag on the thesis this site keeps repeating: the model is becoming a commodity; the leverage is in the harness. The difference is ownership. OpenCLAW and Hermes are open, model-agnostic harnesses you control and can move between. Antigravity is Google's: proprietary, fast, beautifully managed, and locked to Google's model and Google's cloud. Convenience on one side, control and portability on the other. It is worth knowing which trade you are making before you build your week on top of it.

For your engineering team, Antigravity is also a real developer tool · agent-native coding, parallel sub-agents, managed execution. If your devs mention it, they mean the coding platform, not the consumer agent your assistant is using.

Where you'll see it

• Underneath Gemini Spark: every Spark task, schedule, and skill runs on Antigravity.
• In the Gemini API as "Managed Agents," where developers call an agentic workflow the same way they call a chat completion.
• In your engineers' stack, alongside Claude Code and Cursor, as an agent-first IDE, CLI, and SDK.
• In I/O 2026 coverage, usually paired with Gemini 3.5 and Spark as Google's "agentic AI" push.

What to do next

The concept that makes Antigravity click is the harness itself. Read what is a harness for the model-plus-harness formula, then what is OpenCLAW for the open version most operators in this community actually run. And if you want the consumer product Antigravity powers, that is what is Google Spark.