From Coding Companion to Multi-Agent Command Center
Antigravity 2.0 marks a decisive shift in Google’s AI code development tools. Originally pitched as a coding environment, Antigravity is now framed as an “agent-first development platform” for building and managing teams of autonomous AI agents. The new standalone desktop app acts as a command center where multiple agents can run in parallel: one generating code, another assembling brand assets, and a third planning product architecture. Supporting this is an expanded Antigravity SDK, co-optimized with Gemini 3.5 Flash, which Google claims outperforms Gemini 3.1 Pro across most benchmarks and runs significantly faster than competing frontier models. Managed Agents via the Gemini API extend this multi-agent coding platform into the cloud, allowing a single API call to spin up an agent that reasons, uses tools, and executes code in an isolated Linux environment with persistent state across sessions.
Antigravity 2.0 Features: CodeMender, Projects, and Enterprise Workflows
Beyond the core agent orchestration hub, Antigravity 2.0 introduces new workflows aimed squarely at enterprise and long-running development tasks. The platform adds project-based organization, multi-repository support, and the ability to schedule recurring actions—turning what began as a single IDE into a broader multi-agent coding platform. A key addition is CodeMender, Google’s security-focused agent designed to automate patching and code hardening workflows. Antigravity CLI inherits familiar concepts like skills, hooks, subagents, and plugins, while the SDK exposes the same agent harness Google uses internally. Managed Agents in the Gemini API let teams programmatically create agents that can reason, call tools, and execute code over multiple turns without re-initialization. Together, these pieces aim to integrate Antigravity into enterprise cloud environments and workflows, positioning it as a direct competitor to suites like Claude Code and other multi-agent development stacks.
Auto-Update Whiplash: Developers Lose Their IDE Overnight
The Antigravity 2.0 release also revealed the risks of aggressive rollout strategies. On May 19, an automatic update transformed many developers’ existing Antigravity setups into an agent-centric chat experience, stripping out familiar tools such as terminals, file explorers, and the embedded code editor. Source control, local terminals, and remote connections disappeared from the default install, with Google splitting the product into separate downloads and requiring users to install Antigravity IDE to regain full editing capabilities. Within hours, complaints crowded community forums and subreddits, with some developers calling the change a “massive step backward” and accusing Google of shipping non-production-ready changes into live environments. For teams in the middle of active projects, the sudden pivot from a stable coding environment to an agent console underscored how disruptive platform-level changes can be when pushed without opt-in migration paths.
Gemini CLI Replacement and the Open-Source Trade-Off
Alongside the Antigravity 2.0 release, Google triggered further controversy by deprecating the open Gemini CLI in favor of the new Antigravity CLI. The Gemini CLI team announced that, as of June 18, the existing open source development agent will stop serving requests for free users and most non-enterprise customers. Gemini Code Assist IDE extensions and GitHub integrations will also wind down for non-paying users, while enterprises on Standard or Enterprise licenses retain access. Gemini CLI will remain usable only via paid Gemini and Gemini Enterprise Agent Platform API keys, effectively closing off a popular free tool. Developers have criticized the Antigravity CLI replacement for being less open and lacking one-to-one feature parity at launch, even as Google emphasizes its improved support for multi-agent environments and consolidated branding under the Antigravity umbrella.
Performance Showcase and Competitive Landscape
Google is pairing Antigravity’s strategic repositioning with aggressive performance claims. The platform is powered by Gemini 3.5 Flash, which Google says can generate at 289 tokens per second—far outpacing benchmark figures cited for rival models such as Claude Opus 4.7 and GPT-5.5. During a live demo, Antigravity orchestrated 93 subagents to build a complete operating system from scratch in 12 hours, processing 2.6 billion tokens across 15,000 model requests before running Doom on the resulting OS. Antigravity 2.0’s ability to run multiple agents on separate tasks, schedule background jobs, and configure behavior via JSON hooks is clearly aimed at high-intensity, automated development pipelines. These capabilities place it squarely in competition with Claude Code and other multi-agent coding platforms. Yet the backlash over broken setups and tighter access controls suggests Google must balance ambitious multi-agent features with trust, stability, and openness if it wants lasting developer adoption.