We need new interfaces for verification

Learn how cmux runs multiple coding agents in parallel, using devcontainer previews, preview comments for feedback loops, and CI/CD UI testing with screenshots.

Overview

I’m going to be presenting cmux (https://github.com/manaflow-ai/cmux) which is an open source coding CLI manager that lets you run Amp, Claude Code, Codex CLI, Cursor CLI, Gemini CLI, Opencode, and other coding agent CLIs in parallel across multiple tasks. Verification is a big problem when you run a lot of agents in parallel. A human’s time to review AI work is limited, so cmux provides three important features:
1) “Vercel preview URLs” based around the devcontainer.json standard, so you can easily encapsulate your entire development environment
2) “Vercel preview comments” inside each deployment frontend that lets you give feedback directly from your website. Comments automatically spin up additional coding CLIs to autofix.
3) Computer use agents in CI/CD that tests UI changes and takes screenshots to easily compare outputs of different models.
We several more ideas on improving verification, but these three are the ones ready today!

Links

Tech stack