Gemini CLI gets a 30-day migration as Antigravity absorbs the terminal

The loudest Google I/O 2026 announcements were Gemini 3.5 Flash and Gemini Spark, but the more immediate shift for developer workflows came from Antigravity. Google announced the Antigravity 2.0 desktop app, Antigravity CLI, Antigravity SDK, and Gemini API Managed Agents in one package. On the same day, a separate Google Developers Blog post said individual Gemini CLI users would be moved to Antigravity CLI.

The point is not just that Google has another command-line tool. Google describes Gemini CLI as an experiment that brought Gemini into the terminal. Now, it argues, developer demand has moved toward multiple agents splitting work and sharing state. In that world, the terminal cannot remain only a standalone command-line product. It becomes a surface attached to a broader agent platform called Antigravity.

The migration schedule makes that platform logic clearer than the launch copy. According to the official Google Developers Blog notice, Gemini CLI and the Gemini Code Assist IDE extension stop processing requests for Google AI Pro, AI Ultra, and free Gemini Code Assist for individuals users on June 18, 2026. Gemini Code Assist for GitHub also stops accepting new GitHub organization installs after that date, with request processing phased out later. Gemini Code Assist Standard, Enterprise, and Google Cloud-backed usage keep access.

For individual developers, this is a short migration window. For enterprises, there is an exception path. That split is the real message of the announcement. Google is lifting coding agents out of a free terminal experiment and into a platform with shared runtime infrastructure, billing, and organizational control.

Why Antigravity moved beyond the IDE

Google's developer highlights frame I/O 2026 as a shift "from prompt to action." The official highlights post claims Gemini 3.5 Flash beats Gemini 3.1 Pro across most benchmarks and is four times faster than other frontier models. Google connects that speed to the Antigravity harness. Faster models matter most when the workflow is not a single answer, but a repeated loop of tool calls, file edits, test runs, and subagent branches.

Antigravity 2.0 does not keep that loop inside an IDE feature panel. Google describes the new desktop app as an agent-first surface. Dynamic subagents let a main agent define smaller tasks and run them in parallel. Async task management keeps long-running work from blocking the main flow. JSON hooks add policy and control points before and after tool calls. Scheduled tasks let agents run at fixed times without a person manually invoking them.

This is wider than a simple IDE race against Cursor or Claude Code. IDEs still matter, but long-running agents do not naturally live only inside an editor window. Refactoring crosses the terminal and CI. Documentation crosses repositories and issue trackers. Deployment checks touch cloud credentials and sandboxes. Google wants Antigravity to be the shared harness that ties those flows together.

What the Gemini CLI transition says

Gemini CLI appears to have been a successful experiment even by Google's own numbers. The official transition post cites millions of users, more than 100,000 GitHub stars, over 6,000 merged pull requests, and hundreds of contributors. Those are usually growth metrics, not shutdown metrics. But Google uses that success as the reason to absorb Gemini CLI into Antigravity CLI rather than keep it as a separate project.

That is the interesting part. Developers liked the terminal UI, fast release cadence, and community extensibility. Google says the new requirement is multi-agent reality. If one agent refactors in the background, another researches a dependency, and a third runs tests and reviews a patch, then the terminal alone is a weak state manager. Conversation history, permissions, hooks, skills, MCP servers, plugins, and sandbox policies need to be shared across multiple surfaces.

So Antigravity CLI is the successor to Gemini CLI, but the core product Google is building is not the CLI. The core product is the server-side harness. The CLI is the fast operating surface for that harness. The desktop app is the visual coordination surface. The SDK is the programmable surface. Managed Agents are the API product.

Google says Antigravity CLI keeps important Gemini CLI capabilities. Agent Skills, Hooks, Subagents, and Extensions carry forward as Antigravity plugins. In the same migration language, however, Google also says initial feature parity is not one-to-one. Teams that turned Gemini CLI into part of their local development environment will need more than a package replacement.

The SDK looks like an internal API for agent apps

The Antigravity SDK announcement is the more technical signal. The official SDK post introduces google-antigravity, a Python preview library. Developers can use Agent and LocalAgentConfig to open an agent lifecycle, then call built-in tools for file I/O, code editing, shell execution, directory search, image generation, and subagent delegation. Custom Python functions, MCP servers, and agent skills can be added into the same execution pipeline.

The important detail is that Google describes the SDK as agent-friendly. It chose Pydantic V2 models, native Python collections, structured outputs, and explicit naming conventions so the code is readable not only by human developers, but also by coding agents that may generate and edit SDK code inside Antigravity. That is a different posture from a typical SDK announcement. The first consumer is not only the person writing code. It is also the agent that will read, modify, and maintain that code.

Safety defaults are part of the product. The SDK documentation says the default LocalAgentConfig enables built-in tools but applies a confirm_run_command() policy for shell access. Fully autonomous execution requires an explicit setting such as policy.allow_all(). That may sound small, but it is a key default in agent platform competition. Once agents can read files and run commands, permission boundaries and auditability become practical bottlenecks, not side notes.

Google's roadmap points to a remote harness, TypeScript and Go support, Gemma integration, plugins, and deeper observability. The remote harness is especially important because it would turn local agent loops and cloud deployment into one abstraction. The idea is to build an agent locally, then put it into a remote execution environment reachable through an Interactions API. If that roadmap lands, Antigravity SDK becomes less of a local automation library and more of an agent deployment SDK for Google Cloud and Gemini API.

Why the $100 Ultra tier matters

There is also a pricing signal in the announcement. Google said AI Ultra starts at $100 per month and gives Antigravity five times higher usage limits than AI Pro. It also mentioned a $100 bonus credit for users who hit their limit before May 25, 2026.

That can look like consumer subscription news. For coding agents, usage limits are the product experience. Agents consume tokens and tool calls much faster than a person asking occasional questions in a chat box. When subagents run in parallel, tests repeat, browsers and filesystems are used, failed patches are retried, and review loops stack up, quota becomes an invisible wall.

So Antigravity will not compete only on whether Gemini 3.5 Flash is smart. It will compete on how many agent loops a developer can run for the same price, how cheaply failed execution can be absorbed, and how quickly an organization can detect runaway cost. It is not accidental that Google mentions Ultra limits alongside Antigravity. Agent platforms sell execution volume, not just model access.

Community discomfort is product risk

Early Reddit threads in GeminiCLI and google_antigravity showed real discomfort. Some users pointed out that Gemini CLI was open source, active, and already had custom layers built around it. They were unsure whether Antigravity CLI would preserve that shape. Others saw June 18 as a fast date for a core developer tool migration. Posts also asked how Antigravity 2.0, Antigravity IDE, Antigravity CLI, and the existing Gemini CLI differ from each other.

It would be too easy to read this as simple resistance to change. Developer-tool trust comes less from feature count than from path stability. When yesterday's CLI becomes part of a larger platform, request-processing schedules change for existing IDE extensions, and free, individual, and enterprise tiers move differently, developers ask whether they should build more internal automation on top of the tool.

Google's logic is also understandable. As Gemini CLI grew, the expectations placed on one terminal project expanded. Multi-agent work, background execution, hooks, MCP, skills, plugins, organization permissions, remote sandboxes, and API deployment are too much for one standalone project to carry cleanly. Platform consolidation is a technically natural move. The product question is whether users experience that consolidation as predictable.

Google's answer differs from OpenAI and Anthropic

OpenAI is connecting models to workspaces through Codex, the Agents SDK, and managed execution. Anthropic is pushing deeply into developer workflows with Claude Code, Agent SDK, and the MCP ecosystem. Cursor combines an IDE experience with its own model and training loop. Google's Antigravity strategy tries to align all three layers at once: desktop app, CLI, SDK, API, AI Studio, Android, Firebase, and Google Cloud under one name.

The advantage is integration. A builder can bring an AI Studio project into Antigravity, start from the CLI and inspect work in the desktop app, write a custom agent with the SDK, then attach a managed remote execution environment. Inside the Google ecosystem, that is a coherent story.

The weakness is in the same place. Strong integration can reduce the predictability of independent tools and the flexibility of an open ecosystem. That is why this announcement is not automatically good news for Gemini CLI users. Even if extensions, hooks, MCP servers, and skills have migration paths, internal scripts, docs, permission policies, and CI usage patterns still need human verification.

What developers should check now

The first task is to identify which path you are on. If you use Gemini CLI through a personal Google account, free Gemini Code Assist, AI Pro, or AI Ultra, June 18, 2026 is the real deadline. If you use Gemini Code Assist Standard or Enterprise, or access it through Google Cloud, you are not in the immediate shutdown path, but your team still has a reason to test Antigravity CLI and 2.0.

The second task is to separate portable assets. Find where your AGENTS.md files, SKILL.md files, MCP server configuration, hooks, slash commands, and extensions live. Check which permissions they require. Google's migration docs describe a path from Gemini CLI extensions to Antigravity plugins, but some pieces may not map one-to-one. Migration success will depend less on the install command than on these small local configuration files.

The third task is to test cost and execution boundaries. Even if Antigravity CLI is faster and shares a stronger harness, multi-agent execution and scheduled tasks will change usage patterns. Individual developers need to see where quota runs out. Teams need to decide which agents get shell access, what hooks block, and how much execution logging they keep.

The terminal is not disappearing, it is being absorbed

Reading this announcement only as "Google is shutting down Gemini CLI" misses half the story. The bigger change is that the terminal is being absorbed into an agent platform. Developers still want a CLI because it is fast, scriptable, useful on remote servers, and lighter than an IDE. Google knows that, which is why Antigravity CLI exists.

But the CLI no longer stands alone. It connects to a desktop app, Python SDK, Gemini API Managed Agents, and Google Cloud enterprise controls through the same harness. This is Google's view of the next coding-agent layer. It is not just asking a model for code. It is operating multiple agent loops while managing cost, permissions, and state.

That makes June 18 more than a product cutoff. It is the date when a successful community terminal tool is folded into a higher-level platform. If the migration works smoothly, Google gains distribution and infrastructure in the coding-agent race. If portability, quota behavior, feature parity, and early confusion drag on, developers may keep spreading across more independent tools. Antigravity's first test is not a model benchmark. It is whether Gemini CLI users still feel at home in the terminal.