<aside> 👥

Claude Code “Team Agent” in Antigravity

Swarm-style development with tmux (and why it changes your throughput)

Most developers use AI as a single assistant. Team Agent turns it into a parallel delivery system — with you as the architect in the loop.

</aside>

<aside> ⚡

Core thesis

Parallel agents beat sequential prompting.

</aside>

<aside> 🧠

What enables it

Claude Code Team Agent + Antigravity + tmux.

</aside>

<aside> 🎯

Outcome

Faster shipping with clearer ownership boundaries.

</aside>

Overview

This doc outlines a specialized integration of Claude Code’s Team Agent features inside the Antigravity IDE environment.

The goal is to reach a pragmatic form of agentic development: multiple AI models collaborate in parallel while you keep architectural control, review changes, and merge output.

1) The “Team Agent” concept (what it actually is)

<aside> 🧩

In Team Agent mode, Claude Code transitions from “one assistant” to a supervisor + specialists pattern.

</aside>

<aside> 🧭

Lead agent

The interface you prompt. Decomposes the mission, assigns work, and merges results.

</aside>

<aside> 👥

Teammates

Parallel sub-agents (T1/T2/T3…) each running an isolated workstream (DB, API, UI, QA).

</aside>

<aside> 🛰️

Orchestration

Antigravity gives “mission control” visibility; Claude Code drives terminal execution.

</aside>

Practical mental model

• You write one high-level prompt.
• The lead agent turns it into concrete sub-tasks.
• Teammates work in parallel with their own context.
• You perform merge + review before anything becomes “real”.

2) Installation & prerequisites

<aside> 🧱

Expectation setting: Team agents burn tokens faster (multiple contexts). Budget for it.

</aside>

A) System requirements

• Antigravity IDE: version 2026.1+ (sign in with Google for Gemini 3.1 Pro planning capabilities)