Antigravity: The Development Environment That Wants to Eliminate Gravity for Code Writers

When Google announced Antigravity on November 18, 2025, it wasn’t just launching another AI‑powered text editor. It was making a broader claim: that the developer’s role is no longer to write code line by line, but to manage independent agents that write, test, browse, and fix in parallel. The name itself is not random — gravity in traditional development is that accumulated weight of setting up environments, switching windows, and manually debugging framework errors. Antigravity, in Google’s vision, is the engine that lifts you above all of that.

A Different Architecture from the Ground Up

Unlike Cursor, GitHub Copilot, and other tools that add AI on the margin of an IDE originally designed for humans, Antigravity was built on a completely different assumption: the smart agent is the primary engine, and the developer is the supervisor. The platform is a fork of VS Code, but its experience has been reshaped around two main surfaces.

The first surface is Editor View, the traditional development environment with built‑in assistance capabilities — sentence completion, embedded guidance, diff views. The second, and more important, surface is Manager View, which resembles a control room more than a text editor. From here, multiple agents can be launched simultaneously: one agent plans the architecture, another writes code, a third runs tests, and a fourth browses the live application through an embedded Chromium browser. Each agent produces Artifacts — task lists, execution plans, screenshots, browser recordings — rather than raw tool calls. These auditable artifacts are what make working with an independent agent less opaque.

The two modes available in this environment reflect different levels of trust: Planning Mode, where the agent raises full documentation for review before implementing any change — suitable for sensitive repositories — and Fast Mode, which skips documentation and executes immediately for small edits and quick UI tweaks.

From November to April: Rapid Evolution and Sharp Turns

The first five months saw substantial shifts worth tracking:

In March 2026, Google launched AgentKit 2.0, an update that added 16 specialized agents, more than 40 functional skills, and 11 pre‑configured commands covering front‑end, back‑end, testing, SEO, and database management. It became possible to ask Antigravity to restructure authentication, write documentation, and verify payment workflows — and have all three tasks performed in parallel.

That same month, Antigravity was integrated directly into Google AI Studio, the browser‑based platform. You could describe an app idea in text and have the agent build the front‑end and back‑end — including Cloud Firestore and Firebase Authentication — automatically, with no manual setup. Support for multi‑user applications, including instant multiplayer gaming, pushed AI Studio beyond what browser‑based tools were accustomed to.

Almost the same day, on March 19, 2026, Google announced the end of Firebase Studio — a platform it had launched just one year earlier — and directed developers toward two paths: Antigravity for local, agent‑based development, and AI Studio for rapid browser‑based prototyping. New workspace creation in Firebase Studio will stop on June 22, 2026, with full shutdown on March 22, 2027. Developer reactions on Reddit and X were blunt: “Another day, another product Google kills.”

The Quota Issue: When a Free Tool Becomes a Hindrance

The most prominent problem accompanying Antigravity in its current phase is not technical but policy‑driven. The platform launched with a promise of “generous usage rates that refresh every five hours,” but accumulating reports in the Google AI for Developers forum revealed what some developers described as a “97% reduction”: what was 300 million input tokens per week became less than 9 million before hitting the cap. Paid plans start with Pro at $20 per month, going up to Ultra at $249.99 per month, with a pay‑as‑you‑go option at $25 per 2,500 credits.

Comparisons with Cursor — which offers a Pro plan for $20 per month with multiple models — have become frequent in the developer community. The irony is that Google, with its massive cloud infrastructure, faces complaints about tight quotas in a product whose essence relies on agents consuming huge amounts of tokens.

Multiple Models in One Place

What sets Antigravity apart from its competitors is flexibility in model selection. Current support includes Gemini 3.1 Pro in High and Low modes, Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, and GPT‑OSS 120B. This means a developer isn’t locked into a single model — you can route planning tasks to Gemini 3.1 Pro, assign Claude Sonnet 4.6 to write code, and let Gemini Flash handle quick tasks.

The embedded Chromium‑based browser holds a structural advantage: Google owns both Chrome and Antigravity, allowing agents to browse, take screenshots, and run the web app in the same environment without switching. The Knowledge Items feature lets agents retain context and useful code snippets to improve future tasks — as long as the project folder stays in its original location, a limitation documented by InfoWorld journalists as a practical gap.

Support for MCP (Model Context Protocol) arrived in early 2026, opening the door to connect agents with GitHub, databases, and external APIs via the standardized protocol.

What Remains Out of the Picture

The platform is not without documented reservations. An InfoWorld review observed a case where an agent accidentally duplicated an entire section of code, discovered only by chance while working on an unrelated part. Google’s official code‑lab warns that moving the project folder can silently break Knowledge Items with no clear way to repair them.

In November 2025, security‑research company Mindgard documented a security vulnerability that exposed Antigravity users to back‑door attacks; Google announced it was in the process of fixing it. For codebases containing personal data or embedded credentials, verifying the current security status remains essential before any production use.

Market Position and What the Pivot Means

The bigger picture reveals that Google is making a double bet: a free tool to attract developers (Antigravity and AI Studio), and the closure of Firebase Studio before it could settle in users’ minds. Competitors — Cursor, which launched its third version on April 2, 2026 with parallel agent windows, and Claude Code, expanding toward Auto Mode — are not waiting.

The question that remains open in the developer community is not about Antigravity’s technical capabilities, which inspire genuine admiration in the Manager View aspect, but about Google’s reliability as a long‑term partner for a development environment. A product that hasn’t yet reached six months of age, paired with sharply reduced quotas and a $249.99‑per‑month plan, raises legitimate questions about whether “generosity” in quotas is a strategic promise or merely a temporary launch phase.