What Antigravity actually is

Antigravity is a development platform that Google announced on 18 November 2025, alongside Gemini 3. It ships as a desktop application built on the familiar VS Code shape, plus a command-line tool and an SDK that this course never touches. It entered public preview on the day it was announced and, at the time of writing, is free to use with rate limits on the Gemini models behind it.

What makes it interesting to a product manager has nothing to do with code. Three things do.

It works on a folder, not on a chat window

A browser chatbot holds whatever you paste into it and forgets the rest. Antigravity opens a folder on your machine and treats every file in it as context. Your competitor notes, last quarter’s PRD, the support export, the messy interview transcript — all of it is addressable. You point at a file with an @ mention and the agent reads it. When it produces something, it writes a file back into that folder, where you can open it in whatever you normally use.

This is the single largest practical difference. Most of the frustration of using a chatbot for PM work is re-establishing context every session. Here the context is the folder.

Two views, because two different jobs

Google gives it an Editor view — a conventional editor with an agent panel down one side, for when you want to watch and steer — and a Manager view, described as a control centre for orchestrating several agents working in parallel across workspaces, so tasks can run asynchronously.

For a PM, Editor view is where you draft, argue and edit. Manager view is where you send three long jobs off and come back to them. Module 1 spends a lesson on each.

It produces artifacts, not a wall of tool calls

Rather than showing you every command it ran, the agent surfaces what Google calls artifacts: task lists, implementation plans, screenshots, browser recordings. The point is that you can check the work without reading the machinery. A plan you can reject before anything happens is the most useful of these, and it is the subject of lesson 1.4.

Which model is doing the work

Antigravity is not tied to one model. Google’s own Gemini 3.1 Pro and Gemini 3 Flash are the defaults; Anthropic’s Claude Sonnet 4.6 and Claude Opus 4.6 and an open-weights GPT-OSS-120B are also selectable. You will notice the difference on long reasoning tasks more than on drafting.

Where it is weak, honestly

• It is a single-player tool. There is no shared workspace, no comment threads, no review flow. Everything you produce still has to be moved into whatever your team actually uses. If your bottleneck is alignment rather than drafting, this will not fix it.
• It is confident when it is wrong. An agent that writes a plausible PRD section from thin context produces something that reads well and is not true. The whole method in Module 2 exists to force grounding in real files.
• Long autonomous runs drift. The further a task runs without a checkpoint, the more likely you are to get a beautifully executed answer to a question you did not ask. Planning mode is the fix, and it costs a minute.
• It is still a preview. Interface details in this course will age. Where a lesson depends on the exact position of a button, we say so.

Who should not bother

If you write no documents, run no analysis and hand off no specs — if your week is meetings and Slack — the return here is small. If you already have a research assistant and a strong analyst, the gain is narrower than the noise suggests. This course is written for the PM who drafts, cuts and re-cuts documents, and who has data exports sitting in a folder that nobody has looked at.