What inspired us to build MUX?
Y Combinator’s ideas around multiplayer agents got us thinking about a problem we’d run into ourselves. When a friend and I worked on a project together, context often got lost. We ran into incompatible setups and dependencies, and ended up redoing work.
We wanted to make building together smoother: one shared workspace where a team could collaborate with an AI coding agent and keep its work, decisions, and context in sync. That idea became MUX.
What did we learn?
Building MUX taught us that good collaboration depends on clear communication, thoughtful handoffs, and keeping project documentation up to date. We also learned that building with AI is about more than saving tokens. You have to think through the product, give the agent clear instructions, and make sure it’s actually doing what the team asked.
How did we build it?
We built MUX as a team of five, including me, using coding agents such as Claude Code, Codex, and OpenCode throughout development. The project brings the team into a shared coding workflow, where people can contribute direction while an AI agent works on the application.
What challenges did we face?
Our first major challenge was integration. Different teammates used different code patterns and dependency versions. When we merged our work for the first time, parts of the codebase didn’t work together, and we spent days reworking it.
Testing was another challenge. Many AI-generated unit tests passed, but they didn’t reveal problems in the full workflow. We spent more time debugging before realizing our tests weren’t exercising the product end to end. We then designed higher-level tests to stress the complete workflow and catch integration issues earlier.
Those experiences reflected the problem MUX was built to address: teams need to share context and coordinate their work, especially when AI agents are part of the process.
Built With
- agentic
- closed-loop
- docker
- nemotron
- next.js
- nvidia
- postgresql
- python
- react
- supabase
- tavily
- typescript
