Moving one person to a new AI coding assistant takes an afternoon. Moving a team means re-licensing seats, reinstalling editor plugins, re-pointing MCP servers and redoing the security review, all before a renewal date procurement picked.
Google just started that clock for Gemini Code Assist. Since October 9, 2026, nobody can buy Standard or Enterprise, and the end-of-sale notice steers everyone to Antigravity. The full comparison on DevToolLab has screenshots and the complete walkthrough; this is the short version.
No Shutdown Day, Just Renewal Caps
Existing contracts stay supported until they end. What Google capped is how long they can keep rolling:
- June 18, 2026: individual users (free, Google AI Pro, Google AI Ultra) lost the extensions and Gemini CLI.
- September 4, 2026: console purchases closed to anyone without a subscription.
- October 9, 2026: Standard and Enterprise stopped being sold.
- January 31, 2027: last chance for an annual plan to renew or add licenses.
- December 31, 2027: last chance for a monthly plan to renew.
My running example: 40 developers in Austin on Code Assist Enterprise, annual contract up March 1, 2027. That date falls past the January cap, so this term is their final one and new hires need seats by January 31. Updates keep coming until then; version 2.104.0 of the VS Code extension landed October 8.
A Seat Product Versus an Agent Platform
Gemini Code Assist is per-seat completion, chat and agent mode for VS Code and JetBrains, plus helpers built into Firebase, BigQuery, Apigee and a few other Google Cloud products. Gemini models only.
Google Antigravity is built around agents, and you can reach it five ways: desktop app, standalone IDE, editor extensions (VS Code, Visual Studio 2026, JetBrains, Zed, Xcode), the agy CLI, or a Python SDK. An agent plans the task, runs commands inside a sandbox (seatbelt on macOS, bubblewrap on Linux) and pauses for your approval on risky steps.
The bigger shift is models. Gemini Enterprise gives you Gemini 4 Argon, 3.1 Pro and the 3.x Flash line, and admins can opt in to Anthropic's Claude Opus 5.5 and Sonnet 5.5, metered per token. Small Gemini models keep summarizing context in the background even when Claude is picked, and that usage may be billed as well.
Before the Austin team commits, three limits matter. Enterprise sign-in is out of preview only in VS Code and Visual Studio; JetBrains, Zed and Xcode are still preview. The standalone IDE isn't supported in enterprise deployments. And Claude runs only in global, US and EU locations, while Gemini keeps regional residency. The data terms are unchanged: Google doesn't train its foundation models on enterprise prompts or code, and VPC Service Controls still apply.
Seat Math for 40 Developers
Code Assist Standard lists at $19 a seat monthly on a yearly term ($22.80 without one); Enterprise is $45 ($54).
Antigravity has a $0 individual tier with weekly rate limits, and Google AI Pro ($19.99) or Ultra (from $99.99) raise them. Companies don't buy Antigravity directly; they buy Gemini Enterprise Standard or Plus seats, which the pricing page lists "starting at $30 USD per seat per month". Each seat drops $10 (Standard) or $15 (Plus) into a shared credit pool, overages are managed, and a no-seat pay-as-you-go option exists too.
So Austin's $1,800 monthly bill (40 x $45) becomes at least $1,200 (40 x $30) with a $400 credit pool. Add Claude tokens and overages, and the real answer is "cheaper per seat, unknown in total" until you've seen a month of agent usage.
Porting a Gemini CLI Setup to agy
Personal config mostly converts itself. According to Google's migration guide, agy offers to import Gemini CLI settings on first run, agy plugin import gemini converts extensions to plugins, and GEMINI.md plus AGENTS.md are read as-is. If a repo lacks an AGENTS.md, the AGENTS.md Generator gives you one that both tools pick up.
Files committed to a repo are on you:
| Item | Before (Gemini CLI) | After (Antigravity CLI) |
|---|---|---|
| Repo skills | .gemini/skills/ |
.agents/skills/ |
| Home-dir skills | ~/.gemini/skills/ |
~/.gemini/antigravity-cli/skills/ |
| Repo MCP servers | .gemini/settings.json |
.agents/mcp_config.json |
| Home-dir MCP servers | ~/.gemini/settings.json |
~/.gemini/config/mcp_config.json |
| Remote server field |
url / httpUrl
|
serverUrl |
Watch the last row. Per Antigravity's MCP docs, the old field names aren't accepted, so a verbatim copy quietly breaks every remote server. This converter writes the new file with the key renamed and only reads settings.json, so teammates still on Gemini CLI are fine:
import json
import sys
from pathlib import Path
# Gemini CLI settings.json -> Antigravity CLI mcp_config.json.
# settings.json is only read, so anyone still on Gemini CLI is unaffected.
LEGACY_KEYS = ("url", "httpUrl")
def convert(settings_path: Path, out_path: Path) -> None:
servers = json.loads(settings_path.read_text()).get("mcpServers", {})
for name, entry in servers.items():
for key in LEGACY_KEYS:
if key in entry:
entry["serverUrl"] = entry.pop(key)
print(f"{name}: {key} -> serverUrl")
out_path.parent.mkdir(parents=True, exist_ok=True)
out_path.write_text(json.dumps({"mcpServers": servers}, indent=2) + "\n")
print(f"wrote {len(servers)} server(s) to {out_path}")
if __name__ == "__main__":
convert(Path(sys.argv[1]), Path(sys.argv[2]))
On a test repo with one server of each kind (remote GitHub over httpUrl, local Postgres over stdio, an SSE feed over url), plus a Git move of the skills folder:
$ python3 gemini_mcp_to_agy.py .gemini/settings.json .agents/mcp_config.json
github: httpUrl -> serverUrl
incident-feed: url -> serverUrl
wrote 3 server(s) to .agents/mcp_config.json
$ git mv .gemini/skills .agents/skills
$ git status --short
R .gemini/skills/release-notes/SKILL.md -> .agents/skills/release-notes/SKILL.md
?? .agents/mcp_config.json
Stdio entries keep command, args and env intact, and headers is still a valid key. I didn't launch agy (it needs a Google login), so this verifies the files, not a live session. The original post traces each path in that table back to the Google doc it came from.
Five Questions Before Renewal
- When do you renew? After January 31, 2027 means this term is the last, and seat additions stop at that date.
- What editors are in use? A JetBrains-heavy team depends on preview sign-in. Pilot it while the old contract still runs.
- Which Cloud integrations matter? Google hasn't said what replaces Code Assist in Firebase, BigQuery or Apigee. Ask your account rep and get it in writing.
- Do you want Claude? Admin opt-in, per-token billing, three locations only.
-
How does repo config move? A pull request per repo: run the converter, move the skills, and only delete
.gemini/once nobody runs Gemini CLI anymore.
Which Move Fits You
- Solo, free tier: you've already been moved. The individual Antigravity plan is free.
- Code Assist Standard in VS Code: move to Gemini Enterprise Standard before renewal. Full editor support, $11 more per seat, credits included.
- Enterprise on JetBrains: finish the current contract and pilot the JetBrains extension while it's in preview.
- Data that must stay in-country: use first-party Gemini only and keep third-party models off.
- Leaving Google: compare the terminal coding agents before the renewal date, not after.
Wrapping Up
Code Assist still works; you just can't keep buying it forever. Pin down your renewal date, editor mix and Cloud dependencies this quarter, and land the repo changes while the last term is still running.
Top comments (1)
Free users losing the extensions and CLI in June, months before a single paid contract expires, is a wild order of operations. I'd be dying to know if Antigravity actually matches Code Assist feature for feature before anyone moves 40 seats over, or if part of this migration is just accepting less