Google just killed Gemini CLI. In its place: Antigravity CLI, a terminal agent that is not just a rename but a ground-up rethink of what a command-line AI tool should be. Announced at Google I/O 2026, it is the terminal surface of a four-product platform that also includes the Antigravity 2.0 desktop app, the Antigravity SDK, and the Antigravity IDE. The stakes are real: Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests on June 18, 2026, for Google AI Pro and Ultra users, as well as free users of Gemini Code Assist for individuals.
From single-model chat to multi-agent orchestration
Gemini CLI proved the terminal could be an incredible interface for agentic tasks, but developer needs shifted. Teams now require multiple agents communicating with each other to split up work and solve complex problems, which means terminal tools need to share a unified backend with the rest of the workflow. That is the core architectural bet behind Antigravity CLI.
The tool is built in Go, supports asynchronous multi-agent workflows, and shares the same backend as the Antigravity 2.0 desktop application. This unified architecture means any future improvements to the core agents will instantly apply to the terminal experience. No more waiting for backports between surfaces.
The engine under the hood
By default, the CLI connects to Gemini 3.5 Flash (High), which was announced at Google I/O 2026. Google positions this model as running four times faster than other frontier models at the same tier, while outperforming Gemini 3.1 Pro on agentic and coding benchmarks. You can also swap models via the --model flag.
The key architectural concept here is dynamic subagents. In a single-agent setup, one model handles everything sequentially, and the context window fills up with the full history of every step, which degrades reasoning quality on later tasks. Subagents solve this by having an orchestrator agent receive your high-level goal, decide how to break it into subtasks, then spawn specialized subagents, each with its own isolated context window, running in parallel. The main agent can define and spawn specialized subagents on the fly to tackle focused subtasks in parallel, without polluting its own context window, and long-running operations run in the background so neither the main agent nor the UI is blocked.
Don't miss what's next in AI
Join 300,000+ engineers and researchers who get the signal, not the noise.
- Full access to in-depth AI research breakdowns
- Be the first to know what's trending before it hits mainstream
- Daily curated papers, repos, and industry moves