惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

N
Netflix TechBlog - Medium
博客园 - 三生石上(FineUI控件)
Martin Fowler
Martin Fowler
博客园 - 【当耐特】
雷峰网
雷峰网
宝玉的分享
宝玉的分享
IT之家
IT之家
J
Java Code Geeks
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
Jina AI
Jina AI
博客园 - 叶小钗
V
Visual Studio Blog
Engineering at Meta
Engineering at Meta
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
月光博客
月光博客
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
云风的 BLOG
云风的 BLOG
美团技术团队
爱范儿
爱范儿
T
The Blog of Author Tim Ferriss
L
LangChain Blog
U
Unit 42
有赞技术团队
有赞技术团队
博客园_首页

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
My PR Merged Into a Graveyard: On the Rise of Antigravity...
member_26847 · 2026-05-21 · via DEV Community

This is a submission for the Google I/O Writing Challenge


Last week, I got the little green banner. Pull request merged. A small contribution to Gemini CLI — a fix I'd been meaning to submit for months — finally made it in. I closed my laptop feeling quietly satisfied.

Twenty-four hours later, Google announced Gemini CLI is being sunsetted.

I'm not going to pretend my PR was significant. It wasn't. But the timing felt like a metaphor so on-the-nose that I almost laughed. And the more I sat with it, the less funny it became.


What Actually Happened at I/O 2026

At Google I/O 2026, Google unveiled Antigravity CLI — a new, Go-built, agent-first terminal tool that replaces Gemini CLI. The pitch is compelling on paper: faster execution, asynchronous multi-agent workflows, a unified architecture shared with the Antigravity 2.0 desktop app. Your complex refactors run in the background. Multiple specialized sub-agents split up the work. It's Gemini CLI, but grown up.

The hard cutoff lands on June 18, 2026. After that date, Gemini CLI stops serving requests for Google AI Pro and Ultra users. Enterprise and paid API key holders get to keep the lights on a little longer, but the message is clear: the community project is done.

Gemini CLI ended its life with 100,000+ GitHub stars, 6,000 merged pull requests, and hundreds of external contributors. A genuinely impressive open-source run for something barely a year old.

Mine was PR number... well, somewhere in those 6,000.


The Bait-and-Switch Pattern We Keep Accepting

Here's what I keep returning to: Gemini CLI was launched as an open-source project. Google invited the world in. Developers (myself included) filed issues, wrote fixes, argued in discussion threads, and built their workflows around it. The community did what communities do — they made it better than Google could have alone.

And then, when it proved the concept and outgrew its usefulness as a scrappy open tool, Google replaced it with a closed, proprietary successor that ships as a standalone desktop application tied to Google Cloud.

This isn't a new story. We've seen it play out before:

  • Google Reader built a loyal community of RSS power users, then shut down.
  • Google Stadia courted game developers with open promises, then folded.
  • Firebase's open-source adjacency slowly drifted toward lock-in.

The pattern is: open source as acquisition funnel, not as philosophy. Launch something generous, gather contributors, validate the idea, close the door.

What stings about Gemini CLI specifically is that Google's own blog post thanks the community. It cites the 6,000 PRs. It acknowledges hundreds of contributors helped shape the product. And then it announces that all of that community investment is being handed off to a closed platform that, crucially, you cannot self-host in the same way.


To Be Fair to Antigravity

I want to be honest here, because I'm genuinely torn.

Antigravity CLI is technically impressive. The argument Google makes — that your terminal tool and your desktop agent need to share a unified backend to handle multi-agent complexity - is a real argument. When you're orchestrating five sub-agents doing parallel refactors, coordinating through a shared, hardened infrastructure matters. A loose open-source repo maintained by community weekends isn't the right home for that.

The Go rewrite is also a genuine quality-of-life improvement. Gemini CLI, for all its strengths, had latency that you felt. A snappier, more responsive tool isn't a marketing claim; it's real.

And for enterprise users — the ones Google is clearly most interested in — the Google Cloud integration, credential masking, and sandboxing are features that actually matter. A Fortune 500 company isn't going to run a community-maintained TypeScript CLI to push production code. Antigravity CLI is built for the people who pay the bills.

I get it. I just don't have to love it.


What We Actually Lost

When Gemini CLI was open source, I could:

  • Read exactly what it was doing when it touched my codebase
  • Fork it and add behavior my team needed without filing a feature request
  • Audit how it handled my code before trusting it with a production repo
  • Contribute and know that my fix would persist

With Antigravity CLI, I can use it. I can give feedback in a community forum. I can file feature requests and hope someone at Google agrees with me. That's a fundamentally different relationship with a tool.

The dev community spent the better part of a decade learning to be suspicious of black-box AI tools that touch their codebases. Gemini CLI, by being open, offered an answer to that suspicion. Antigravity CLI doesn't.

For a tool that wants to "orchestrate agents across your entire workflow" — touching your files, your Git history, your credentials — the closed-source decision feels like exactly the wrong call.


The Larger Shift I/O 2026 Is Signaling

Step back and look at what Google announced this year as a whole:

  • Gemini Spark is a "24/7 personal agent" that accesses your Gmail, Calendar, and Tasks
  • Antigravity 2.0 is a desktop app that steers agents across your projects
  • Managed Agents in the Gemini API provisions full agent infrastructure for you
  • Antigravity CLI handles your terminal

There's a coherent vision here, and it's not an open ecosystem. It's Google as the substrate. Every agentic workflow you build runs on infrastructure Google controls, through interfaces Google designs, against models Google hosts. The Antigravity SDK lets you "customize the agent," but you're still hosting it on Google's harness.

This is Google's answer to Microsoft's Copilot+ and Anthropic's Claude Code: a vertically integrated agentic platform. And it's probably going to be very good. That's sort of the problem.

The best agentic tools are becoming the ones most deeply locked into vendor ecosystems. The open alternatives - still worth rooting for, but always a step behind - can't match the integration depth that comes with a single company owning the whole stack.


What This Means for You (And for Me)

If you're a developer, the practical advice is boring: Antigravity CLI will probably be the better tool by August. Learn it. Use it. Enjoy the async workflows.

But keep asking the uncomfortable questions:

  • What happens to your workflows when Antigravity gets deprecated in 2028 for Supersymmetry CLI?
  • Where does your code go when it passes through a closed agent harness?
  • Who owns the institutional knowledge these tools build about how your codebase works?

The Google I/O keynote ended with the line: "We're bringing the models, infrastructure, and platforms together across Google to help you build anything you can imagine."

That sentence is doing a lot of work. Across Google. Building on Google. Anything you can imagine - so long as Google's infrastructure agrees.


My PR #19881 is sitting in a deprecated repo. The fix still works - remove unsafe type assertion suppressions in error utils, if you want to go look. The code is there, archived and frozen.

And somewhere in a Mountain View conference room, someone is presenting a roadmap slide that doesn't have Gemini CLI on it anywhere.

That's fine. That's how this industry moves. I just think it's worth naming it clearly, before we all sign up for the next open-source launch and start filing pull requests again.


Follow me here on dev.to if you want more takes on developer tooling, open-source sustainability, and the occasional misadventure in AI-assisted coding.