Google Antigravity

You will like the Agent Manager: several agents running in parallel across projects while a browser subagent opens the result and checks it, on a free Individual plan with weekly rate limits. The cost is calm, and that is rare. What wears on you is the preview feel: the app wants to run things, and features move under you between weeks.

Pick it for greenfield sprints where three agents building three screens at once is the point and a broken week costs nothing. Pick Cursor for the codebase you maintain daily, where one agent that edits well and stays out of the way beats a swarm you supervise.

The risk is reach. An agent holds the editor, the terminal and a live Chrome window at once, and there is no container sandbox in the capability list, so a wrong shell command lands on the machine you are sitting at and a wrong click lands in a browser session that may be logged in. Three surfaces, one permission model, and the person watching is the only boundary.

Keep the browser profile separate from your own and leave auto-run off for a week. What it does right: New Worktree mode runs the agent in an isolated git worktree, so a bad diff stays off your branch.

Antigravity publishes no benchmark, which leaves the architecture to judge. 1. Scope: a Project defines which folders and repositories an agent may touch, so the retrieval boundary is declared rather than discovered. 2. Verification: agents emit artifacts, implementation plans, walkthroughs and screenshots, for a human to review, which moves verification from the tool log to a document. 3. Autonomy: a /goal command runs a task to completion without intermediate approval.

The design bet is that reviewing artifacts scales better than reviewing actions, which is plausible and unmeasured. The observation stands: a screenshot is evidence of rendering, not of correctness, and a plan is evidence of intent, not of execution.

The company is Google, so runway is moot and roadmap is everything, and Google's developer-tool roadmap has historically been decided by a reorganization rather than a market. Monetization is the tell: higher quotas come from Google AI Pro at $20 and Ultra at $100 or $200, consumer bundles that make the IDE a subscription feature rather than a product with its own line. A feature does not get a P&L; it gets a review.

The pivot risk is consolidation, not death: folding into another surface, a rename, or losing the models Google does not own. No acquirer applies. Position: adopt for individuals, keep team workflow portable, assume one rename.

The demo shows one developer supervising a swarm, which excites a VP and worries a CTO. Procurement sees an Organization plan billed by consumption through Google Cloud, with no published per-seat figure, so I cannot multiply anything by sixty; I can only forecast, and consumption forecasts for agents are wrong by the third month. The CLI has a headless mode, which gives CI a path. Identity, audit and retention are not described on the pricing page.

Onboarding is a Google login, which is the easy part. Not yet: a fixed seat price or a hard consumption cap, plus retention terms, before a pilot goes beyond ten people.

Closed box. The IDE gives me no key field and no source to read, and a local model means pointing GOOGLE_GEMINI_BASE_URL at a Gemini-shaped shim I write myself, so misbehaviour ends at a support forum. The agy CLI at least accepts a Gemini API key, MCP servers live in a mcpServers JSON block with a serverUrl for remote ones, and a Python SDK scripts the whole thing, so I drive it from code I own.

Grudging respect for the SDK; it is the only part that treats me as a programmer. Nothing here is mine, and nothing survives Google losing interest. I would script it, not depend on it.