Isomux Backstory

In January, I read Steve Yegge's Gas Town post where he shows the following hierarchy:

Steve Yegge's 8 levels of coding agent usage, from Y/N confirmations in a single chat (level 1) to an orchestrator managing a fleet of Claude Code instances (level 8)

I was at Level 5. The post warns you not to try to jump straight to Level 8, so I looked for one thing I could pick up from it to help me up the hierarchy.

That one thing was tmux (short for Terminal MUltipleXer), which did help with managing a few Claude Code sessions at once. But I didn't want to spend all day every day looking at terminals. So I got a bit creative and made a wrapper that was cuter to look at, with isometric graphics. Hence, Isomux (ISOmetric MUltipleXer):

That started a pattern: evolving Isomux to solve pain points in my environment.

For example, another pain point was having to manage a mix of local and remote sessions.1 For the former, I had to keep my laptop open; for the latter, I had to ssh all the time.

So I moved Isomux to a server. This way:

  • All my devices see the same agents and conversations.
  • All my agents outlive my laptop and see the same environment.

Over time, I developed Isomux into a meta-harness:

Three nested boxes. The innermost is the model. Around it is the harness, which asks how we make the most out of a model, with Claude Code and Codex as examples. Around that is the meta-harness, which asks how humans make the most out of agents, with Omnigent, qm, and Isomux as examples.

I started Isomux on March 27, 2026, and I built it from inside it since that day.2 In fact, the only times I've used anything else since then is when I accidentally bricked Isomux during development and had to use "bare bones" Claude Code to unbrick it.

And yes, it helped me go up the Steve Yegge hierarchy:

My weekly commit count from Nov 2024 to Oct 2026, banded by stage of Yegge's hierarchy: Level 4 with Cursor as daily driver, Level 5 with Claude Code from Nov 2025, and Levels 6-8 with Isomux from late Mar 2026. Weekly commits climb from single digits to about 650 in the week of Oct 19, 2026

Office Metaphor

In January, I also saw Vibecraft, which shows Claude Code sessions as animated characters. That planted the idea of anthropomorphized agents.

The office metaphor comes from trying to make agents more accessible to non-technical folks. The idea is that talking to an agent should feel like talking to a coworker - something relatable to everyone.

Each agent has a customizable name and look and sits at a desk. You see who is working, who's sleeping, and who needs you (hand raised) at a glance.

Timeline

  • Mar 27, 2026: First commit.
  • Mar 28: Multi-device and multiplayer (in the same VPN). The same office, live on laptop and phone. Messages carry the sender's name, so several people can talk to the same agents.
  • May 9: Agent-to-agent messaging.
  • May 10: Multi-provider. Codex agents, side by side with Claude Code agents.
  • May 14: Member accounts with invite links (multiplayer no longer restricted to the same VPN).
  • Aug 24: Turned Isomux into a company and launched my first product, Hosted Isomux (isomux.app).

By May, Isomux had all four of multi-user, multi-agent, multi-device, and multi-provider in the same conversation; as of October, I haven't found another product that does.3

Building in public

Footnotes

  1. Tasks like model training can only be done remotely. ↩︎

  2. Here's Isomux on its first day, next to how it looked in August. ↩︎

  3. The Isomux dates are from its git history. The closest products I found: Databricks Omnigent (June 2026) has multi-user, multi-device, and multi-provider, with agents working across sessions on its roadmap. Devin (April 2026) and Amp (July 2026) added multiplayer sessions, and Claude Code added cross-session messaging in v2.1.224 (August 2026). ↩︎

Stay in the loop

I'd love to tell you when I publish a new post.

Get notified when I write about DS&A or software engineering. Unsubscribe anytime if it's not your vibe.

Prefer RSS? Subscribe via rss.xml.

Want to read more? Here are other posts:

Loop Engineering

Loop Engineering

IsomuxAI

What I learned from moving Isomux development from work sessions to a continuous loop of agents.

Research 2.0

Research 2.0

ResearchAI

Building Big O Speedrun in two days: how agents discover, verify, and broadcast new research results, and my agentic workflow.

Nil Mamano

Computer scientist, software engineer, author.