This is a submission for the Google I/O Writing Challenge
If you have been using Gemini CLI to run agentic tasks from your terminal, you have until June 18, 2026 to migrate. After that date, requests stop being served for Google AI Pro, Ultra, and free-tier users. No warnings. No grace period. Just errors.
That deadline is the most important thing to come out of Google I/O 2026 that most developers are not talking about. Everyone is writing about Gemini 3.5 Flash benchmarks and the Antigravity desktop app. Meanwhile, a large chunk of the developer community is sitting on a tool that is about to break their workflows.
So let us talk about what Antigravity 2.0 actually is, what changed, and what you need to do before June 18.
What Google Actually Announced
At I/O 2026, Google announced Antigravity 2.0 as a full agent-first development platform. It ships in four parts:
- Antigravity 2.0 : a rebuilt desktop app
- Antigravity CLI : a Go-based terminal tool (the Gemini CLI replacement)
- Antigravity SDK : for building custom workflows
- Managed Agents API : spins up an autonomous agent with a single API call
The headline for most people was the desktop app and the new Gemini 3.5 Flash model powering it. The headline that actually matters for your day-to-day workflow is the CLI transition.
Google's reasoning makes sense when you read it plainly. Gemini CLI was built for single-agent, terminal-based tasks. But developer workflows have shifted. You now need multiple agents communicating with each other, splitting work across a unified backend. Gemini CLI was not built for that. Antigravity CLI is.
What Actually Changed
The Binary
Gemini CLI ran as gemini. Antigravity CLI runs as agy. That is not just a rename. The entire thing was rewritten in Go, which means no runtime dependencies and noticeably faster response times.
If you have Gemini CLI invocations sitting in shell scripts, GitHub Actions, or Dockerfiles, every single one of those needs to be updated. For most cases it is a straightforward find-and-replace, but Dockerfiles with GEMINI.md references need explicit file mapping.
The Config Files
This is where the real migration work lives.
| Old (Gemini CLI) | New (Antigravity CLI) |
|---|---|
GEMINI.md |
AGENTS.md |
.gemini/skills/ |
.agents/skills/ |
Your project config and your skills both move. The good news is that both binaries can coexist on the same machine during the migration window. gemini and agy run separately with separate config trees and separate auth flows. You can keep running gemini on specific tasks until June 18 while you test the migration on others.
The Architecture Shift
The biggest actual change is how Antigravity CLI handles complex tasks. Gemini CLI locked up your terminal session while it worked. Antigravity CLI uses an asynchronous background processing model. You can kick off a large-scale refactor or a multi-source research task and your terminal stays free.
Multiple agents run in parallel, with the practical ceiling sitting at around four or five before performance starts to degrade. The multi-agent system is configured through AGENTS.md, a markdown file where you define agent roles, communication patterns, and orchestration rules in plain text. If you have used Claude Code, you will recognize the pattern immediately. It works similarly to CLAUDE.md. No complex orchestration code, Just markdown.
The SKILL.md System Is the Part Worth Getting Excited About
This is getting buried under the broader Antigravity coverage, so I want to spend real time on it.
In Antigravity, a Skill is a directory-based package. Each skill contains a SKILL.md file and optional supporting assets like scripts and templates. The design is intentionally simple: markdown and YAML, formats every developer already knows.
Here is what a basic skill structure looks like:
.agents/skills/
└── my-skill/
├── SKILL.md # Instructions and rules
└── scripts/
└── run.sh # Optional supporting script
What makes this powerful is the distinction between Skills and Rules. Rules in Antigravity are persistent. They are always on, functioning like a system prompt that applies to everything the agent does. Skills are on-demand. They only load when triggered, which means you can build specialized capability extensions without bloating every single agent interaction.
Think about what this means practically. You can define a skill for how you want your API routes structured, one for your testing conventions, another for your deployment checks. The agent loads the relevant ones when it needs them. Your AGENTS.md defines the team. Your skills define how each agent does its job.
For a frontend developer, this is a real productivity shift. You are not re-explaining your conventions to an AI every session. You write them once in markdown, put them in .agents/skills/, and the agent follows them consistently across every project that uses that skill.
Who Is Actually Affected and What to Do
Free and Google AI Pro/Ultra users: You must migrate before June 18. This is mandatory. Start now.
Enterprise users on Gemini Code Assist Standard or Enterprise licenses: Your access remains unchanged. Migrate when it suits you.
Enterprise Gemini Code Assist for GitHub: Same deadline applies. No new installations on GitHub organizations after June 18, and requests stop being served in the weeks following.
The migration itself is not complicated. Google estimates it takes under ten minutes for most setups. Here are the steps that actually matter:
# 1. Install Antigravity CLI
# Follow instructions at antigravity.google
# 2. Import your existing plugins
agy plugin import gemini
# 3. Rename your config file
mv GEMINI.md AGENTS.md
# 4. Move your skills directory
mv .gemini/skills/ .agents/skills/
# 5. Update CI/CD scripts
# Find and replace 'gemini' with 'agy' across your shell scripts and workflows
Then test one real workflow before you close out Gemini CLI entirely. That is the step most people skip, and it is the most important one.
One real frustration worth flagging: the /usage command in Antigravity CLI only updates after you quit and reload, not live. Developers have reported running out of quota mid-task with no warning during long agentic workflows. Google has acknowledged this as a known issue with no fix date. If you are running tasks that go longer than twenty to thirty minutes, keep that in mind and plan accordingly.
My Take
Antigravity 2.0 is a real product shift, not a rebrand. The move from single-agent to multi-agent, the asynchronous architecture, the Go rewrite, the AGENTS.md configuration system, these are meaningful changes that reflect how serious Google is about owning the agentic development space.
The SKILL.md system in particular has the kind of depth that takes a few sessions to fully appreciate. The separation between persistent rules and on-demand skills is a thoughtful design decision that does not get enough attention.
That said, this migration is mandatory and the community reaction has been honest. Plenty of developers on GitHub explicitly preferred Gemini CLI and wanted it to stay. The plugin ecosystem in Antigravity CLI is still maturing, and the live quota tracking issue is a real operational gap. Antigravity CLI is not feature-equivalent to Gemini CLI on day one, and Google should say that more plainly than it has.
What matters most is this: if you wait until June 18 to start, you are setting yourself up for a bad day. The migration is short. The rollback is free until the deadline. Run one real task through agy this week and see where things break. That is the only way to know what your specific workflow actually needs.
The June 18 deadline is not a recommendation. Set the reminder now.
I am Tobore, a full-stack developer and Best Author on DEV for React writing. I built the Delta State Health Information System and I write about tools that actually change how developers work. Find more at dev.to/toboreeee.
Top comments (0)