From AI Coding Tool to Multi-Agent Development Platform
Antigravity began life as an AI-assisted coding environment, but Antigravity 2.0 marks a decisive strategy shift. Google now describes it as an “agent-first development platform” designed to develop and manage teams of autonomous code agents rather than a single assistant living inside an editor. The new standalone desktop app operates as a central command center, where developers can orchestrate multiple agents working in parallel – for example, one focused on code generation, another on product architecture, and a third on assets or documentation. This reframing places Antigravity alongside emerging multi-agent development stacks, not just traditional AI coding tools embedded in IDEs. Technically, the upgrade leans on Gemini 3.5 Flash, co-developed as Antigravity’s primary workhorse model and tuned for coding and agentic workloads, with Google claiming it outperforms Gemini 3.1 Pro and runs significantly faster than rival frontier models on key benchmarks.
Inside Antigravity 2.0: Command Center, CLI, SDK and Managed Agents
Antigravity 2.0 reorganizes Google’s AI coding tools into a multi-layer platform. At the top is the desktop command center, which can coordinate several agents in parallel, group conversations into projects, span multiple repositories, and schedule routine background actions. For terminal-first workflows, Google is consolidating its tooling under the new Antigravity CLI, effectively deprecating the Gemini CLI while preserving features such as skills, hooks, subagents, and plugins. An Antigravity SDK exposes the same agent harness that powers Google’s internal products, co-optimized with Gemini 3.5 Flash for higher throughput and agentic reasoning. In the cloud, managed agents in the Gemini API can be spun up with a single call into persistent, isolated Linux environments where tools can run, code can execute, and state survives across sessions. Together, these pieces turn Antigravity from an IDE into an end-to-end multi-agent development and deployment suite.
CodeMender, AI Security Patching and the Enterprise Pitch
Beyond orchestration, Google is using Antigravity 2.0 to push deeper into production workflows and AI security patching. A new capability dubbed CodeMender focuses on automated remediation: scanning codebases, proposing fixes, and applying security patches via autonomous code agents rather than one-off suggestions. Combined with JSON-configurable hooks and scheduled tasks, Antigravity can continuously run background checks, file updates, and dependency maintenance, positioning it as a proactive AI security layer over existing repositories. Enterprise appeal is further sharpened by integration with the Gemini Enterprise Agent Platform and tighter ties to Firebase and Android tooling, signalling that Antigravity is meant to plug directly into cloud governance and CI/CD pipelines. However, analysts note many enterprise buyers may delay large-scale adoption until Google publishes clearer terms around governance, deployment controls, and operational security for autonomous agents operating across production environments.
A Rocky Rollout: Breaking IDE Workflows Overnight
The strategic pivot to an agent-first platform was overshadowed by a rollout that alienated many existing users. On May 19, an automatic update replaced working Antigravity setups with Antigravity 2.0, effectively gutting the familiar IDE-like environment. Developers opened their machines to find core tools – code editor, local terminals, file explorers, and remote connections – replaced by what felt like a bare chat interface focused on autonomous agents. Google had quietly split the product into three downloads, but the auto-install only delivered the new agent app; teams needed to manually fetch the separate Antigravity IDE to restore their previous workflow. Forums and social channels quickly filled with frustration, with some calling the agent-first shift a step backward and criticizing the lack of rollout controls. For a platform pitching reliability and autonomy, breaking thousands of active development environments in a single push raised immediate concerns about production readiness.
Competitive Positioning: Going After Claude Code and Beyond
Strategically, Antigravity 2.0 is Google’s clearest answer yet to competitors like Claude Code and Codex. By turning Antigravity into a unified brand that covers editor-based coding, terminal workflows, and autonomous code agents, Google is moving from experimental tools toward a cohesive AI coding platform. Performance claims are aggressive: Gemini 3.5 Flash, which powers Antigravity 2.0, is said to beat Gemini 3.1 Pro on coding and agent benchmarks and reach 289 tokens per second, far above the rates publicly cited for rival models. Live demos of 93 subagents collaborating to build a full operating system – complete with running Doom – highlight the vision of large-scale multi-agent development. Yet the backlash around the forced update underscores a key differentiator for enterprises: not just raw capability, but governance, rollout safety, and fine-grained control over autonomous code agents operating in real production systems.