Google debuts Antigravity 2.0 to take on Cursor

Last year, Google launched Antigravity as a direct shot at Cursor and the wave of AI-first coding tools. At Google I/O 2026, the company decided a code editor plugin was thinking too small.

Enter Antigravity 2.0, now a full-blown standalone desktop application across macOS, Linux, and Windows. Google is expanding the tool beyond pure software engineering into broader knowledge work.

Agents running agents

The biggest shift here is orchestration. Instead of just chatting back and forth with a single AI in an IDE sidebar, Antigravity 2.0 lets you direct multiple agents and run tasks asynchronously.

You can now set up custom subagent workflows and schedule automated background jobs using standard cron syntax. There are also new JSON hooks for automation pipelines, alongside new slash commands like /goal and /grill-me, the latter designed to interrogate your logic before running code.

Google also baked in native voice commands, mirroring what it recently rolled out to Gmail and Google Docs.

Under the hood, Antigravity 2.0 runs on Google's new Gemini 3.5 Flash model. Google even noted that the model was co-developed using Antigravity itself.

Alongside the desktop app, Google introduced an Antigravity CLI, an SDK, and native API access. Developers using the older Gemini CLI are being nudged to migrate to the new tool immediately.

Google Cloud enterprise customers get built-in connectors, and AI Studio now features custom agent templates with local export options. Google is even piping Antigravity capabilities into Google Search to generate real-time UIs and mini-apps on the fly.

Pricing also got a shakeup. Google launched a new $100-per-month AI Ultra plan that delivers 5x higher usage limits than the Pro tier. Meanwhile, it shaved its top-tier Ultra plan from $250 down to $200 a month, offering 20x higher limits to match Anthropic and OpenAI's high-tier tiers.

Why it matters

If you are a developer already tied into Android, Firebase, or Google Cloud, Antigravity 2.0 is worth testing immediately. Being able to kick off background agent tasks with cron and have subagents run integration tests while you work on other tasks saves serious context-switching overhead.

From an industry perspective, Google is making a smart structural pivot. Rather than getting bogged down fighting specialized IDEs on their home turf, Google is reframing the agent as a standalone desktop operations hub, treating coding as just one of many tasks an agent can handle.

The catch? Orchestrating multiple dynamic subagents running in the background eats tokens fast. Even with the top-tier plan dropping to $200, letting autonomous agents churn unattended could lead to surprising billing spikes if guardrails fail.