Codex adds Symphony multi-agent and Computer Use cross-application

OpenAI developer account @OpenAIDevs posted continuously on May 12 to promote Codex’s two major recent features: 「Symphony」, a multi-agent collaboration framework, and 「Computer Use」, cross-application operating capabilities. These two features let Codex evolve from an “AI assistant inside the IDE” into an “agent that can operate autonomously across apps, across tasks.” According to @OpenAIDevs’ posts, each of the two features received 700–850 likes.

Symphony: a dedicated Codex agent for every opened task

Symphony is the “Codex orchestration spec” that OpenAI open-sourced on April 27. The core idea is to turn a project management dashboard (such as Linear) into a control plane for coding agents:

In the project management dashboard (such as Linear), each “opened ticket” automatically gets paired with a Codex agent

Autonomous agent execution: pull the ticket, write code, run tests, submit PRs, and continue until merging

Humans shift into “reviewers”: review PRs after the agent finishes the work

The spec is open-sourced, so any team can adopt it

After OpenAI publicly released and internally piloted Symphony, PR submission volume increased by 6x. For development teams, Symphony clearly splits “human labor vs agents” into assignments—humans define the tickets and review, while agents handle execution and testing.

Computer Use: operate across apps, but not take over the Mac

Computer Use is one of the core capabilities introduced in Codex’s April 16 “Codex for almost everything” update. According to a conversation video between @OpenAIDevs and engineers @AriX and @romainhuet, this feature allows Codex to:

Operate across applications on the user’s Mac—click, input, and switch windows

But “not take over” the computer: users can still work on other tasks at the same time

Background execution: Codex handles its workflow, while humans do their own things

Applies to: automating cross-tool tasks, handling repetitive GUI operations, and integrating non-API services

The combination of Computer Use and Symphony lets Codex move from an “AI assistant inside the terminal” to a full-stack agent that can navigate the desktop environment and connect tasks across tools.

Competitive landscape: Codex vs Claude Code—clear battlefield

Chain News observation: this week, Codex and Claude Code are both pushing functions in similar directions, making the competitive landscape clearer:

Multi-agent collaboration: OpenAI Symphony (Linear ticket-driven) vs Anthropic Agent View (managing multiple sessions from a single interface)

External tool integration: Codex adds Computer Use (cross-GUI applications) vs Claude Code connects via MCP (file system, git, and custom tools)

Distribution mode: Codex integrates Codex CLI, VS Code, IDE plugin, and desktop apps; Claude Code is CLI-first, with desktop integration in progress

Fundamental differences between the two approaches: Codex embraces the existing desktop environment and aims to directly control the user’s existing tools; Claude Code builds an agent-specific execution environment, isolating with worktree and standardizing external integrations via MCP. The former lowers the user learning curve; the latter emphasizes security and controllability.

For enterprise users, the choice depends on two dimensions: the team’s existing workflow (heavy reliance on Linear workflows or heavy IDE integration), and how acceptable “AI autonomous desktop operation” is. In the next 6–12 months, the two are expected to positively clash on “enterprise integration SDKs” and “multi-agent collaboration specs.”

Events to watch next include: Symphony open-source community adoption rate, whether third-party project management tools (Jira, Asana, ClickUp) follow up with Symphony integration, and Codex Computer Use’s expansion plans to Windows and Linux platforms.

This article, “Codex adds Symphony multi-agent and Computer Use cross-application,” first appeared on Chain News ABMedia.