From Coding Companion to Multi‑Agent AI Platform
With the Antigravity 2.0 update, Google is recasting its AI coding environment as a full multi‑agent AI platform rather than a single smart editor. Announced during Google I/O, the release introduces a command center that can orchestrate several agents in parallel, manage longer coding and automation chains, and organize work into projects spanning multiple repositories. Antigravity now bundles editor surfaces, terminal-style execution, and agent-building tools under one product name, positioning it as a unified suite of Google AI development tools instead of scattered experiments. Under the hood, Antigravity 2.0 is powered by Gemini 3.5 Flash, with Google touting aggressive token throughput and live demos of large agent swarms building complex systems. Managed agents in the Gemini API add an execution layer for background jobs and evaluations, while an SDK gives developers a standard way to wire these agents into real workflows.
Automatic Rollout Chaos: When the IDE Vanishes Overnight
Hours after the launch, the Antigravity 2.0 update triggered widespread outrage because of how it shipped, not just what it promised. An automatic update landed on developer machines and effectively gutted the existing IDE experience. Terminals, file explorers, and in-editor tools many relied on for daily work disappeared, leaving behind what felt like a chat-first interface tuned for agents, not humans. Google simultaneously split Antigravity into separate downloads: the new Antigravity 2.0 app focused on AI agent orchestration, while the traditional Antigravity IDE became an optional install users had to discover and fetch manually. Active projects were suddenly unusable until developers reassembled their environments. Forums and social channels filled with complaints that the rollout favored product vision over operational readiness, with many arguing such breaking changes should have been opt-in, staged, or at least clearly flagged before deployment.
Goodbye Gemini CLI, Hello Closed‑Source Antigravity CLI
The Antigravity 2.0 update also marks a decisive shift at the command line, where Google is retiring the open-source Gemini CLI in favor of a new, closed-source Antigravity CLI. Google frames the move as consolidation, promising better support for multi-agent environments and carrying over core features such as skills, hooks, subagents, and plugins. But the migration is far from gentle. On June 18, Gemini CLI and its associated IDE extensions will stop serving requests for most Pro, Ultra, and free individual users, effectively forcing a developer API migration onto the new tool. Only customers on certain standard or enterprise licenses retain seamless Gemini CLI access, and the open-source GitHub repository gives way to a thin Antigravity CLI page with just documentation and change logs. With no 1:1 feature parity at launch and the loss of source transparency, many developers see this as a step backward in openness and control.
Access Tiers, Enterprise Controls, and Friction by Design
Beyond UX whiplash, Antigravity 2.0’s access model deepens the divide between individual developers and enterprise customers. On June 18, Gemini CLI access for most non-enterprise users ends unless they route traffic through paid Gemini and Gemini Enterprise Agent Platform API keys. In practice, that means teams without enterprise credentials or explicit billing arrangements are nudged toward the new Antigravity CLI, with fewer options to stay on tools they trust. At the same time, Google is pitching Antigravity as a serious enterprise platform, bundling governance-friendly features, managed agents in isolated Linux environments, and higher-usage plans like the AI Ultra tier at USD 100 (approx. RM460) per month. For large organizations, these controls and limits help standardize AI development. For smaller teams, they feel like gates: stricter quotas on old tools, more friction to experiment, and a forced climb up Google’s access ladder.
New Power Tools: Multi‑Agent Command Center and CodeMender
Amid the controversy, Antigravity 2.0 does deliver substantial new capabilities aimed at serious software work. The multi‑agent command center lets developers coordinate swarms of agents on parallel tasks, schedule recurring actions, and run long-lived workflows that outlast any single prompt or session. In Google’s own demos, dozens of subagents collaborated across billions of tokens to assemble complex systems, showcasing how the platform can handle large-scale automation. Alongside that, the release introduces CodeMender, a security patching tool designed to plug into these workflows and continuously apply fixes across codebases. Combined with the managed-agent execution layer and the Antigravity CLI’s keyboard-first control surface, the suite aspires to be an end-to-end environment for building, running, and securing autonomous AI agent teams. The tension, for now, lies in whether these gains outweigh the pain of breaking changes and tightening access for the broader developer community.