Antigravity, Claude Code, and Codex in one system: who does what

Using Antigravity, Claude Code and Codex in one system is how a single person can divide work among three programming agents from three companies (Google, Anthropic, OpenAI) with clear roles: one agent as coordinator keeps the big picture, the execution agents fix code in the right project and on the right machine they're assigned to, and everyone reads the same rulebook and status file. This post does not rank the three tools, does not promise to save time or money because OPCAIOS hasn't measured, and the role division is our convention, not any company's recommendation. The comparison table comes from official documentation from each side and from --help of each CLI, read on 03/11/2026. Dividing work between two agents already exists in the post using Codex and Claude Code at the same time; this post focuses on where Antigravity differs and its role in the trio.

One coordinator agent, execution agents, one rulebook and one shared status file. Diagram: OPCAIOS.

How are coordinator and execution agents different?

A coordinator agent is the one that holds the big picture of the whole system: reads status, checks services, delegates work, tracks delegated tasks and records results; an execution agent is the one that takes one specific task in one specific project, fixes code, runs tests, commits and reports back. This is OPCAIOS's convention, not any company's feature: all three tools can fix code and run commands, so the roles are set by the owner. The reason to split roles:

1. One place holds shared context: knows which service is down and who is holding what task.
2. One owner per project: reduces work of two agents editing on top of each other.
3. Easy debugging: immediately know which session, which machine made the fix.

Comparison table of three tools' roles per official documentation

The table below records only facts seen in documentation or --help on 03/11/2026; the last column is how OPCAIOS is currently using it, not a quality judgment:

Criterion	Google Antigravity	Claude Code	Codex
Surface	Antigravity app, Antigravity IDE, CLI agy, utilities for other IDEs	Terminal, VS Code, JetBrains, desktop app, web	ChatGPT app (desktop, mobile, web), Codex CLI, IDE utilities, Codex Cloud
Model	Gemini, Claude, GPT-OSS by package (/docs/models)	Claude; CLI and IDE utilities support additional third-party providers	OpenAI's models
Instruction file	AGENTS.md or GEMINI.md, cumulative from every level	CLAUDE.md; reads AGENTS.md when no CLAUDE.md	AGENTS.md, from user directory down to project
Run non-interactive	agy -p	claude -p	codex exec
Separate working directory	App supports worktree; sub-agent can run on separate worktree	Flag -w, --worktree	Flag --worktree
Role at OPCAIOS	Coordinator	Execution	Execution, review

The foundational post What is Antigravity has a section introducing the app, IDE and CLI if you're new to this tool name.

How does Antigravity differ from Claude Code and Codex?

Antigravity differs from the other two tools in five points that affect how you wire the system, per documentation at antigravity.google read on 03/11/2026. No point says Antigravity is stronger or weaker; they just change how you set rules and divide work:

1. Cumulative rules: per /docs/rules, Antigravity reads AGENTS.md or GEMINI.md from the working directory and parent directories up to the repo root, cumulative global rules, then merges them all. Claude Code prioritizes CLAUDE.md.
2. Multiple model families: Gemini, Claude or GPT-OSS by package, see post choosing model in Antigravity.
3. Built-in sub-agents: /docs/subagents lists research, browser and self; browser runs a browser in sandbox, called via /browser.
4. Team of agents: /teamwork command for multi-day projects, only in paid tier.
5. Login terms: /docs/faq says using third-party software (naming Claude Code) to access Antigravity using an Antigravity account is against the terms, and wrapping agy -p inside a third-party agent is not supported on personal plan.

The fifth point matters most: do not have Claude Code or Codex call Antigravity as a "worker" using a personal account. The three tools coordinate through shared files, each tool logs in with its own company's account.

Why is Antigravity assigned the coordination role in the OPCAIOS system?

Antigravity is assigned the coordination role in the OPCAIOS system because the way Antigravity organizes work fits with monitoring multiple areas at once, not because it has been measured to be better. According to documentation read on 03/11/2026, the application can run multiple agent conversations in parallel, the main agent can call multiple sub-agents at once and track them in a separate table, and every package has scheduled tasks. The coordinator agent's work includes:

1. Check infrastructure and read the status file at the start of the session.
2. Assign work by writing tasks, projects, machines, and main agents into the status file.
3. Monitor results, update incidents and logs.

Claude Code and Codex also have sub-agents, scheduled runs, and non-interactive modes, so switching roles is still possible. What must be kept is that only one coordinator agent runs at a time, otherwise the status file will have two conflicting plans.

How should AGENTS.md be set up as a shared file for three agents?

AGENTS.md shared is a rules file that all three tools can read, but each tool finds the file in its own way. According to documentation read on 03/11/2026: Codex reads AGENTS.md from the user's Codex directory then from the project root downward; Claude Code reads AGENTS.md when there is no CLAUDE.md, or when CLAUDE.md imports it; Antigravity reads AGENTS.md or GEMINI.md and combines them, with global rules at ~/.gemini/GEMINI.md or ~/.gemini/AGENTS.md. Suggested layout:

1. Common rules (what to read at session start, what to write at session end, who edits which project) write once in AGENTS.md.
2. CLAUDE.md only imports @AGENTS.md plus Claude Code's own section.
3. GEMINI.md keep very short, since Antigravity combines it with AGENTS.md; if the two files say different things the agent receives conflicting commands.

Detailed syntax is in the post What is AGENTS.md; Antigravity's rules in .agents/rules/ directory format is in the post rules and workflows in Antigravity.

Why "one project edits on one machine only"?

"One project edits on one machine only" is a rule that each code repository is edited and deployed from exactly one machine, even though the system has many machines and many agents. This rule came from a real mishap in OPCAIOS: when one agent deployed a website from a Windows machine, the comparison showed over 1,500 files different only because of line ending type (Windows uses CRLF, macOS and Linux use LF). Fake changes hid real changes, and another agent might accidentally commit a whole batch. With three agents, the risk is higher because the coordinator agent and execution agent are often on different machines. Rules applied:

1. Each code repository clearly writes its managing machine in AGENTS.md and the status file.
2. Only assign editing work on that repository to the agent on that exact machine.
3. To change machines, change it in the status file first.

The standard way to normalize line endings with .gitattributes is in the post Multiple AI sessions editing one project.

How does the shared status file help hand off between coordinator and executor?

The shared status file is the handoff channel between the coordinator agent and the executor agent, because the three tools do not share memory and should not call each other using personal accounts. One line assigning work writes all four things:

1. Work to do, one sentence description.
2. Project and machine being edited.
3. Main agent and time of assignment.
4. Status: in progress, done, stuck, with time of actual check.

For example, not a real person: the coordinator agent sees a registration page error, writes one line assigning to Claude Code on the machine holding the website repository; the Claude Code session reads that line, fixes it, commits the right file, writes "done" with test results; the next coordinator session checks the real page then closes the incident.

OPCAIOS Experience

OPCAIOS is the website opcaios.org run by one person using AI agents. These things are real, verified on 03/11/2026:

1. The main Mac has Antigravity application version 2.28.x, Antigravity IDE version 2.5.x and CLI agy version 1.4.12, used in parallel with Claude Code and Codex.
2. Antigravity keeps the coordinator role: managing infrastructure, assigning and tracking tasks. It has also performed some administrative features for the website, for example tracking the Dream 100 audience on Facebook and X, and editing the website menu.
3. Website code has a rule that edits and deployments only happen from a different machine (a Windows machine), after seeing more than 1,500 different files due to line ending styles.
4. There are over 400 saved Antigravity conversations stored locally on the machine (counted by directory, not read by content).
5. GEMINI.md at the user level has only one line: always reply in Vietnamese. General rules are in AGENTS.md in the user's home directory and in the Obsidian note vault.
6. The Antigravity application declares 14 MCP servers; hooks are only used to report session status, no command-blocking hooks yet; the only global workflow is an empty template; no knowledge sections have been seen being created.
7. Not yet using Antigravity SDK, not using Remote Control, not using /teamwork, and productivity of this role-splitting approach has not been measured.

Frequently Asked Questions

Do I need to use all three tools?

No. The three tools are OPCAIOS's choice, not a requirement for doing OPC. One agent used skillfully, with AGENTS.md, with habits of committing the right files and reading diffs, is enough for most solo work. Add tools only when there's a specific reason, like needing a different plan's rate limits or having someone from another vendor review. Each tool added is another subscription cost and another set of rules to keep in sync.

Can I use an Antigravity account with Claude Code?

No. The /docs/faq page of Antigravity read on 03/11/2026 states that using third-party software, naming Claude Code specifically, to access Antigravity with an Antigravity account violates the terms and may result in account lockout. Google suggests using an API key from Gemini Enterprise or Google AI Studio if you want a third-party agent to use Gemini. The page also says wrapping agy -p in a third-party agent is not supported with the personal plan. This is a description of the terms, not legal advice.

Does Antigravity read AGENTS.md like Codex does?

It does read it, but merges it differently. According to /docs/rules read on 03/11/2026, Antigravity looks for AGENTS.md or GEMINI.md in the working directory and parent directories up to the repo root, plus global rules, then merges them all into the prompt. It does not read rules in subdirectories below the working directory. Because they stack, if GEMINI.md and AGENTS.md say different things, the agent receives both, so keep common rules in a single file.

Should the coordinator agent edit code itself?

It can, but only if that work is recorded in a status file like any other task, and only on the machine allowed to edit that project. At OPCAIOS, Antigravity has directly edited administrative features and website menu. The risk is the coordinator agent gets used to editing on its own machine, breaking the rule "one project one machine". When doing it, the coordinator agent records itself as the agent owner of that task.

Where should someone new start combining three tools?

Start with rules, not tools. Write one common AGENTS.md, one short status file, and set which machine each project edits on. Use one agent until the habit of reading status and keeping a log runs smoothly, then add a second agent in a review role that only reads, finally separate the coordinator role. If you want guidance building a multi-agent process with controls, see OPCAIOS's AI Application Training course.

Source: Google Antigravity documentation at antigravity.google (/docs/rules, /docs/subagents, /docs/teamwork, /docs/agent, /docs/models, /docs/plans, /docs/faq), Claude Code documentation at code.claude.com (overview page, memory), Codex documentation at learn.chatgpt.com (overview page, AGENTS.md), agy --help 1.4.12, claude --help 2.1.317, codex --help 0.162.1; read on 03/11/2026. Not verified: effectiveness of the coordinator and executor role split, cost of running three plans in parallel, Antigravity CLI worktree capability; features may change with version. Experience is OPCAIOS's.