Coinbase discloses AI programming tool Mux: engineers are shifting from "code writers" to "AI agent commanders"

robot
Abstract generation in progress

BlockBeats News, May 12 — Coinbase recently disclosed the development status of its internal AI programming tool Mux, stating that as AI Coding Agents become more widespread, engineers’ roles are gradually shifting from traditional code implementers to “multi-agent orchestrators.”

Coinbase stated that although AI programming tools like Cursor, Copilot, and Claude Code have significantly improved the efficiency of individual engineers, the traditional development process remains serial, with engineers able to handle only one branch or task at a time.

To address this issue, Coinbase’s internal engineers developed a multi-agent programming tool called Mux. This tool allows multiple AI Agents to work in parallel within separate Git worktrees, branches, and terminals, while engineers are responsible for unified review, feedback, and code merging. Data shows that by April 2026, Mux has rapidly spread within Coinbase:

—Over 600 users;

—Covering 461 code repositories and 10 organizations;

—A total of 5,068 PRs generated and merged;

—Mux users average 3.5 times more PRs merged than regular engineers (39.6 vs. 11.4).

Coinbase pointed out that AI is transforming the core capability model of software engineering. In the future, the truly important skills will no longer be just “coding speed,” but problem decomposition, system design, code review, and agent collaboration management skills.

The article also emphasized that Mux initially was just a side project for individual engineers, but thanks to Coinbase’s early development of internal LLM Gateway, agent infrastructure, and an open experimental culture, the tool eventually achieved spontaneous diffusion.

Coinbase believes that AI is also changing the internal tool development logic of “self-developed vs. purchased.” Internal tools that previously took months to procure and evaluate can now be prototyped within days by individual engineers with the help of AI.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin