From AI Coding Assistant to Multi-Agent Command Center
Antigravity 2.0 marks Google’s most aggressive attempt yet to move beyond a traditional AI coding assistant. Originally framed as an IDE-style environment, the product is now a standalone desktop hub for coordinating teams of autonomous code agents. Instead of a single model responding to prompts, developers can spin up multiple specialized agents: one writing application code, another drafting brand assets, and a third planning product architecture or refactoring repositories. Under the hood, Antigravity 2.0 runs on Gemini 3.5 Flash, which Google claims beats Gemini 3.1 Pro on coding and agentic benchmarks while outperforming competing frontier models on speed. The update also adds an agentic layer that supports asynchronous task execution, parallel subagents, and tighter context management so long-running projects do not stall or overflow context windows. In effect, Antigravity 2.0 shifts the developer’s role from direct editor of code to manager of coordinated, semi-autonomous workflows.
New Tools: CLI, SDK, Managed Agents and CodeMender
To support multi-agent development at scale, Google is surrounding Antigravity 2.0 with a wider toolbelt. The Antigravity CLI gives terminal-focused developers a way to spin up agents, preserving familiar concepts like subagents, skills, hooks, and plugins from the older Gemini tooling. A new SDK exposes the same agent harness Google uses internally, optimized for Gemini 3.5 Flash, and lets teams programmatically define and orchestrate autonomous code agents inside their own applications and services. Managed Agents in the Gemini API extend this model into the cloud, where a single call can provision an agent inside a persistent Linux environment with stateful files and multi-turn sessions. Alongside these capabilities, Google introduced CodeMender, an automated security patching system designed to let agents identify and fix vulnerabilities as part of continuous workflows. Together, these pieces recast Antigravity as a platform for end-to-end, multi-agent development rather than just an AI coding assistant embedded in an editor.
Breaking Workflows: Auto-Update Fallout and Gemini CLI Deprecation
The ambitious shift to multi-agent development has come at a steep usability cost for many developers. An automatic update pushed before the Antigravity 2.0 announcement effectively gutted the existing editor: terminals, file explorers, source control panes, and remote connections vanished overnight, replaced by a chat-centric agent UI. Google simultaneously split Antigravity into separate downloads, forcing users who depended on the integrated development environment to fetch a distinct Antigravity IDE just to recover basic coding capabilities. In parallel, the company is deprecating the Gemini CLI and steering users toward the new Antigravity CLI, consolidating its AI coding assistant branding under a single product name. While this unification clarifies Google’s platform story, it also breaks established workflows and automation scripts built around the old CLI. For many teams, the transition feels less like an upgrade and more like a disruptive reset that arrived without sufficient warning or migration tooling.
Access Limits and Enterprise Readiness Questions
Despite the broadened toolset, access to Google’s AI coding assistant ecosystem is narrowing for many individual developers. As of June 18, Gemini CLI access is being removed for most users, with capabilities now increasingly tied to Antigravity 2.0, enterprise offerings, and paid Gemini API access. That effectively shifts the most powerful multi-agent development features toward organizations able to standardize on Google’s enterprise AI tools, while hobbyists and smaller teams see their previous setups degraded or paywalled. On paper, Antigravity integrates with the Gemini Enterprise Agent Platform and Google Cloud surfaces, hinting at a future where governance, security, and deployment are centrally managed. In practice, buyers still lack detailed policies and reference architectures for topics like permissions, auditability, and safe rollout at scale. Until those governance and security terms are clearly defined, many enterprise engineering leaders are likely to treat Antigravity 2.0 as a promising experiment rather than a primary development platform.
Competing with Claude Code and Codex in the AI Dev Stack
By reframing Antigravity 2.0 as an agent-first platform, Google is overtly targeting rivals like Claude Code and OpenAI’s Codex-based tools. The standalone desktop app, cross-platform support on macOS, Linux, and Windows, and Gemini 3.5 Flash’s reported performance gains position Antigravity as a serious contender in the AI-assisted development market. Where competitors emphasize single assistant interactions inside editors, Antigravity leans into multi-agent development: grouping work into projects, spanning multiple repositories, and scheduling background tasks that agents execute autonomously. New AI subscription plans starting at USD 20 (approx. RM92) per month reinforce Google’s intent to commercialize the stack, from desktop workflows to cloud-managed agents. The strategic bet is clear: future software teams will orchestrate fleets of autonomous code agents rather than rely on one-off code completions. Whether developers embrace that vision may depend less on benchmark wins and more on how quickly Google can repair trust after the rocky Antigravity 2.0 rollout.