AI Coding Tools Have a Six-Month Half-Life: The Gemini CLI Lesson

On May 19, Google announced that Gemini CLI will stop serving requests on June 18, 2026 for Google AI Pro and Ultra subscribers and for free users. Standard and Enterprise customers keep access. The replacement is Antigravity CLI: a closed-source Go binary distributed inside Google's Antigravity 2.0 agent-development platform. Gemini CLI was six months old. It was Apache 2.0 TypeScript with volunteer contributions on GitHub. It is getting killed.

Most coverage will tell you this is migration news. Bye-bye Gemini CLI, hello Antigravity CLI. That is true. It is also the least interesting thing about what just happened.

The interesting thing is the timeline.

How long did Gemini CLI last?

Six months. That is how long a major vendor's flagship coding CLI lasted before the same vendor killed it for the next thing.

Three days ago we wrote about the five-vendor convergence on the agent-CLI surface: Anthropic shipping Claude Code, OpenAI shipping Codex CLI, GitHub shipping Copilot CLI, xAI shipping Grok Build, Google shipping Gemini CLI. The convergence had its first casualty within 24 hours of us naming it.

This isn't coincidence. Convergence is what happens when many vendors agree on a product category. Convergence also costs something. The cost is products getting killed when the category moves faster than they were built for. Six months from launch to sunset is what that cost looks like at one vendor. It will look similar at the others.

What did Google actually concede?

Google's own framing is candid about why. From the migration announcement:

Two phrases worth holding up.

"Early days of 2025" arrived inside six months of the product shipping. That's the velocity admission. Whatever the language sounds like, the company is calling a less-than-a-year-old product an early-days product. The reason isn't engineering taste. The reason is that the category moved.

"Single product" means the company is choosing concentration over breadth. They are not maintaining two coding CLIs in parallel. They are picking one and putting their weight there. That's a structural choice about how to compete in the agent surface, and it explicitly accepts the cost of orphaning the previous product.

This is not anti-feature framing. It is a velocity statement about the category. Google didn't say Gemini CLI was bad. They said it was built for an earlier reality that already is not theirs.

Why does open-to-closed matter?

There's a second cost in this transition that the migration framing buries.

Gemini CLI was open source. Apache 2.0 TypeScript. Volunteers reviewed PRs and shipped features. The agent-CLI surface that opened across these five vendors was at least partially open. Anyone could read the code, vendor a copy, fork it if they needed to.

Antigravity CLI is closed source. The Antigravity SDK is a Python wrapper around a bundled Go binary. The volunteers on the gemini-cli GitHub are publicly asking whether their contributions will get absorbed into the closed enterprise product they did not sign up to build. Google has not answered.

This isn't an OSS-advocacy point. It's a structural one. The agent-CLI category opened in an open-source moment and is consolidating closed at the same time it is consolidating around fewer products per vendor. If your reason for picking an open-source agent CLI was portability or vendor-independence, that bet just got harder for Gemini CLI users. The next vendor's transition will land in roughly the same place.

Who keeps access on June 18?

The third cost is access tiering.

On June 18, Google AI Pro and Ultra subscribers lose Gemini CLI access. Free users lose it. Only Standard and Enterprise license customers retain access.

The agent-CLI category just got harder for the solo builder to reach as a first-class tool from this vendor. Whatever Antigravity's mission statement means, the distribution of access to the new CLI is narrower than the old one's was.

Sit with the three costs together. Six-month product half-life. Open-source replaced by closed-source. Solo and Pro tiers lose access while business tiers keep it. These aren't three separate observations. They are three faces of the same thing: vendors are consolidating the agent-CLI surface as fast as they can, and the consolidation is sorting access along business-tier lines. The category isn't just five vendors agreeing on a product. It is also five vendors agreeing on who that product is for, which increasingly is not the lone developer with a $20 subscription.

What should you build against, then?

If you architected something around Gemini CLI in the last six months, you have 28 days to migrate. That's the immediate decision.

The bigger decision is what you architect around next.

The honest answer is to build with the surface, not on it. The portable layers in 2026 are the model (where the actual intelligence lives) and the protocol layer (MCP, where interoperability lives). Vendor-specific surface affordances of any one CLI (its flags, its config format, its distribution mechanism) are a moving target. The vendor that ships the surface today is the vendor most likely to ship a replacement six months later.

This is continuous with what we wrote in subscription shock vs usage drift: the variables in AI coding are not the model capabilities, which are improving steadily. They are the surface and the pricing, which are changing every quarter. Plan against the stable things. Watch the moving things. Move when they move.

We grounded this earlier in the three layers of collaboration. The agent-to-agent layer (which is what these CLIs serve) is the one moving fastest because that's where vendors are still figuring out what the surface should look like. The team that treats their build practice as bound to one vendor's view of that surface is buying a product with a known short half-life. The team that treats vendor surfaces as interchangeable carriers for the underlying model and protocol is more durable to whatever the next sunset looks like.

This is not a counsel of despair. It's a posture. You can use any one vendor's CLI as your daily tool. You just shouldn't be surprised when the tool stops being maintained, and you shouldn't build infrastructure that assumes it will.

What this isn't

We're not saying Google was wrong to sunset Gemini CLI. The multi-agent-reality argument is honest. Antigravity CLI is probably the better tool for the workflows Google is now optimizing for. The criticism is not of the choice. It is of the speed at which the choice was made and the cost the speed transfers to the builders who have to keep migrating.

Six months from "early days of 2025" to "we sunset it" is a velocity statement about the category, not about Google specifically. The next vendor will do the same. Our prediction: within 90 days.

The surface is a moving floor

The agent-CLI is becoming infrastructure at the same speed that vendors are still figuring out what infrastructure means. Six-month half-lives. Open-to-closed transitions. Tier-based access splits. These are the operating conditions for AI building right now.

The team that treats them as conditions, not as crises, ships through them. The team that bets hard on one vendor's coding CLI in May 2026 is paying for a product that may not exist in November.

The surface is a moving floor. The floor will move again.