Most AI coding tools still treat the human as the one doing the work, with the model whispering suggestions into an editor. Antigravity inverts that: the agent is the primary actor, and the developer becomes a reviewer who approves plans and inspects results. The unlock isn't a smarter autocomplete — it's that agents are given the same surfaces a human uses (editor, terminal, and a real browser) and are expected to verify their own output instead of handing back raw diffs.
What Sets It Apart
- Two modes for two mindsets: an Editor View for hands-on coding with an agent sidebar, and a Manager surface for spawning and watching multiple agents run asynchronously across workspaces.
- Agents act across editor, terminal, and browser, so a task like "build this feature and confirm it works" can include actually clicking through the running app.
- Output comes back as Artifacts — task lists, implementation plans, screenshots, browser recordings — rather than opaque commits, which makes reviewing async agent work tractable.
- Model optionality: defaults to Gemini 3 Pro with generous preview rate limits, but also runs Claude and GPT-OSS, so you aren't locked to one provider.
Who It's For
Great fit if you want to delegate larger, multi-step tasks and supervise them, or run several agents in parallel while you focus elsewhere — and you value being able to audit what an agent actually did. Look elsewhere if you mainly want fast inline completions in your existing editor, distrust autonomous agents touching your terminal and browser, or need a stable production tool today: it shipped as a free public preview and the agent-first workflow is a real adjustment from assistant-style tools.
