Agent Manager with Google Antigravity/Real Fleet Projects
Lesson 5 / 6 · 17 min · 0 / 4 exercises
Course contents
Concept · 4 hands-on exercises

Gemini CLI, Antigravity, or the API: Choosing Google's Agent

Know which Google agent fits which job as of November 2026: the Antigravity App and CLI on personal accounts, Gemini CLI for Code Assist licenses, the SDK and the API agent for code you ship, and what Jules' shutdown means for you.

Google ran several coding agents side by side, and in 2026 it consolidated them. If you learned on Gemini CLI or Jules, or you're about to build an agent into your own product, this lesson sorts out which tool is meant for which job as of November 2026.

Gemini CLI: still shipping, for a narrower audience

Gemini CLI is Google's open-source terminal agent, and it still ships: v0.68.0 landed on November 10, 2026. What changed is who it's for. Its README says free individual access and Google AI Pro and Ultra subscriptions "moved to Antigravity CLI on June 18, 2026." Gemini CLI is now designed for Gemini Code Assist Standard and Enterprise licenses and Code Assist for GitHub on Google Cloud, where existing access is unchanged, and the README says it's "Not tuned for the Gemini 4 series."

On a personal Google account, move to agy. The migration is light. GEMINI.md and AGENTS.md work unchanged, and ~/.gemini/GEMINI.md is still read. Extensions become plugins; the migration guide's command is agy plugin import gemini. MCP servers move out of ~/.gemini/settings.json into a standalone mcp_config.json, where a remote server's URL key becomes serverUrl. Workspace skills move from .gemini/skills/ to .agents/skills/.

Spot the problemsExercise 1 of 4

A teammate moved their Gemini CLI MCP servers into Antigravity. Find the lines that won't work or shouldn't ship.

.agents/mcp_config.json0 of 2 found

Jules: winding down

Jules, Google's asynchronous cloud coding agent, is shutting down. Google's notice of October 29, 2026 let sessions continue until November 11, made Jules read-only until the end of the year, and says "Jules will wind down on December 31, 2026." The notice says its lessons are shaping Google products, "including our agent-first platform, Google Antigravity." Export before the deadline. Jules Tools exports your settings, scheduled tasks, connected repositories, memories, setup scripts and every session to a local folder, while your code, branches and pull requests stay in GitHub. The export includes your environment variables with their secret values, so keep that folder private.

The closest thing to Jules inside Antigravity is remote agents: sessions that run in an isolated, Google-managed environment and keep working when your laptop sleeps. As of November 2026 they're rolling out gradually to Google AI Pro and Ultra subscribers on personal accounts. Start one from a New Remote Project in the App, or with agy --remote-agent on CLI 1.4.12 or later. Remote sessions can't see your local files, so you attach repositories to the remote project.

Quick checkExercise 2 of 4

Every night, Jules fixed your repo's dependency alerts and opened PRs. What's a realistic replacement plan?

Building agents into your own product

Two options exist for agents that run inside your code rather than in your editor: the Antigravity SDK, a Python library on the same harness as the App, and the managed Antigravity agent on the Gemini API. Both are as documented in November 2026.

# pip install google-antigravity   (Python 3.10+)
# export GEMINI_API_KEY="your_api_key_here"
import asyncio
from google.antigravity import Agent, LocalAgentConfig

async def main():
    config = LocalAgentConfig()
    async with Agent(config) as agent:
        response = await agent.chat("What files are in the current directory?")
        print(await response.text())

if __name__ == "__main__":
    asyncio.run(main())
Antigravity SDKAPI agent (antigravity-preview-09-2026)
Where it runsYour machine or your servers, on the App's harnessA Google-hosted Linux sandbox per environment
ModelGemini models, or local models through LiteRT or an OpenAI-compatible serverBuilt on Gemini 3.8 Flash; configurable through agent_config
ControlCustom Python tools, skills, MCP, and policies for tool approvalAGENTS.md and .agents/skills/ mounted into the sandbox, remote MCP servers, a max_total_tokens budget
BillingGemini API per-token pricing, or Vertex AIPreview; pay-as-you-go on the underlying model's tokens and the tools it uses
Fill the blanksExercise 3 of 4

Start the managed Antigravity agent in a fresh Google-hosted sandbox.

interaction = client.interactions.create( agent="", input="Run the tests and summarize the failures.", environment="", )
Sort itExercise 4 of 4

Which Google tool fits each situation?

Card 1 of 5

A product feature that lets customers ask an agent to analyze their files

One more on-ramp
Google AI Studio is powered by the Antigravity coding agent, and its Export to Antigravity integration moves a whole AI Studio project, with its context, into local development.
Check your understanding
Q1. You're on a personal Google AI Pro account and want a terminal agent. Which should you install as of November 2026?
Q2. What's true about Jules as of November 12, 2026?
· Work through the 4 remaining exercises above (stuck? each one can show you the answer).
· Score 100% on the quiz.
Vocabulary in this lesson

Plain-English definitions from the AI Coding Dictionary.