Google Kills Gemini CLI: What the Forced Switch to Antigravity CLI Means for Your CI/CD Pipelines

On June 18, 2026, Google switched off Gemini CLI — the 105,000-star open-source terminal agent that millions of developers had been using to automate Gemini-powered workflows since Google I/O 2025. There was no soft deprecation, no grace period on the free tier, and no automatic migration. Any terminal script, CI/CD pipeline, or cron job that called the gemini command simply stopped working. The replacement arrived the same day: Antigravity CLI, a closed-source Go binary built on the same agent harness that powers Antigravity 2.0, launched at Google I/O 2026. The switch is instructive not only as a migration task but as a signal about the direction of the entire AI developer tools industry.

What Happened on June 18

Gemini CLI and Gemini Code Assist IDE extensions stopped serving requests for Google AI Pro, Google AI Ultra, and free individual tier users on June 18, 2026 — the deadline Google had announced at I/O 2026. Enterprise users on Gemini Code Assist Standard or Enterprise and developers with paid Gemini or Gemini Enterprise Agent Platform API keys were not immediately affected. But for the majority of individual developers using the free or consumer paid tiers — the bulk of the open-source community that made Gemini CLI's GitHub repository one of the fastest-growing developer tools in 2025 — the shutdown was immediate and unconditional.

Five Breaking Changes That Caught Teams Off Guard

The five changes that caused the most real-world breakage: The binary name changed from gemini to agy — every script calling gemini failed with command not found. The MCP server configuration format changed from inline settings.json to a separate mcp_config.json file with renamed fields — a silent failure that produces no error output, making it especially dangerous in automated pipelines. Authentication in CI/CD shifted to requiring the AGY_TOKEN environment variable or the --auth-token flag with a service account token. The daily request quota dropped from approximately 1,000 per day to approximately 20 per day, resetting weekly rather than daily. And the codebase moved from an Apache 2.0-licensed JavaScript project on GitHub to a closed-source Go binary — eliminating the ability to audit, fork, or self-host.

The Open-Source Reversal

The open-source dimension of this transition is generating the most sustained developer frustration. Gemini CLI launched in mid-2025 as an Apache 2.0-licensed Node.js project. It accumulated 105,000 GitHub stars in under a year, driven by developer adoption precisely because it was inspectable, forkable, and self-hostable. Anthropic's Claude Code and Microsoft's extensions remain either open or partially open. Google's decision to replace Gemini CLI with a closed-source binary is the sharpest break from that ecosystem expectation, and developers who had built internal tooling on top of the source tree — writing custom middleware, extending the agent loop, or deploying the CLI inside air-gapped environments — have no upgrade path other than a complete rewrite against the new binary's interface.

The 98 Percent Free Tier Cut

The rate limit change is the second flashpoint. Gemini CLI's free tier offered approximately 1,000 requests per day — generous enough to run meaningful automation, run agents in development environments, and prototype multi-step workflows without incurring costs. Antigravity CLI's free tier is approximately 20 requests per day — a 98 percent reduction — and the reset cadence changed from daily to weekly. The practical implication is that developers who built lightweight CI scripts running on every pull request, or who used the free CLI to run nightly analysis jobs, have been cut off entirely from their current usage patterns. A migration to Antigravity CLI is not merely a binary name swap: it is a forced evaluation of whether the use case justifies a paid tier.

What Antigravity CLI Actually Brings

Setting aside the friction of migration, Antigravity CLI has genuine technical improvements over its predecessor. The Go binary delivers sub-millisecond cold start times, eliminating the Node.js runtime dependency that created version conflict headaches on developer machines and in container environments. Memory usage is significantly lower. The agent architecture underneath has been redesigned for multi-agent orchestration: Antigravity CLI is not a conversational AI assistant running in a terminal — it is a harness for composing teams of specialized agents that can operate in parallel, hand off work, and maintain state across long-running sessions. The shared harness between Antigravity CLI and Antigravity 2.0 desktop means workflows built in the terminal are portable to the desktop environment. For teams that can absorb the migration cost and the rate limit reduction, the architectural foundation is meaningfully more capable than what Gemini CLI offered.

The Larger Pattern

The Gemini CLI shutdown is the second major AI developer tool disruption in three weeks. On June 1, 2026, GitHub replaced Copilot's premium request billing with token-metered AI Credits, triggering a wave of developer reports about unexpected cost spikes — one Pro+ subscriber exhausted 8 percent of their monthly credit allowance in just two hours. The pattern is the same in both cases: AI tools that had been priced as flat-rate developer utilities are being repriced as metered compute infrastructure, and the recalibration is happening faster than developer workflows can adapt. The reason is structural. Agentic AI tools — those that make multiple model calls per user interaction, maintain long sessions, and pull in large context windows — cost dramatically more per interaction than the chatbot-style tools they replaced. The economics of offering them at flat-rate subscription prices or with generous free tiers no longer holds as usage shifts from casual assistance to deep workflow integration.

What to Do Right Now

If your team has scripts or pipelines calling gemini, the migration has three concrete steps. Install Antigravity CLI with the standard install command and verify that agy --version responds correctly. Rename all invocations of gemini to agy in shell scripts, Makefiles, GitHub Actions workflows, and CI configuration files. Migrate the MCP server configuration: move any MCP server definitions from settings.json to a new mcp_config.json file using the updated field names — this is the silent failure that breaks pipelines without producing any error output, so test it explicitly with a pipeline run that exercises the MCP tool path. If you were relying on the free tier for meaningful automation, evaluate the paid tier against your actual usage before assuming the migration is cost-free. The AGY_TOKEN authentication change is straightforward but must be set as an environment secret in your CI platform.

Bottom Line

The Gemini CLI shutdown is a useful forcing function. It surfaces a question every engineering team integrating AI developer tools into their workflows needs to answer: are the tools you are building automation around stable infrastructure, or are they services whose economics are still being discovered? The honest answer in mid-2026 is that AI developer tools are still early-stage infrastructure — capable, accelerating, and genuinely valuable, but subject to pricing recalibration, architecture changes, and deprecation on timescales that production infrastructure cannot accommodate without planning. Teams that treat AI tools as load-bearing infrastructure with no exit strategy are the ones getting hurt most by transitions like this. The engineers navigating these changes most smoothly are the ones who wrapped their AI tool integrations behind an internal abstraction layer — a thin adapter that lets them swap the underlying CLI without rewriting every pipeline that depends on it.