So we’ve all heard about that parallel universes theory that physicists get all excited about, right? Well, I doubt many of them have used tmux, so how much can they really tell us about actually living in multiple realities at once?
I've been doing this dance for almost 15 years now. Fifteen. That's longer than some JavaScript frameworks have existed. Hell, that's longer than some JavaScript frameworks have existed twice.
The Problem With Having Just One Brain
You know that feeling when you're cooking dinner, answering a text, helping with homework, and trying to remember if you fed the cat? That's what modern agentic software development feels like now, except the cat is a Claude Code instance and it's definitely hungry.
Our computers are phenomenal at context switching. They do it thousands of times per second, juggling processes like a circus performer on espresso. But us? We're still using tools designed for when having one terminal window open was considered fancy.
I once tried to explain my workflow to someone and they looked at me like I was describing how to perform surgery with kitchen utensils.
The Dance of Many Projects
Let’s try again, here's how it works:
One Session Per Project. This is my cardinal rule. Each project gets its own named session: protocollie, sublayer-gem, apm_macos, the-12th-estate like little kingdoms in terminal land.
Context is everything, and your brain (or at least my brain) works better when it knows exactly where it is. When I switch to the protocollie session, suddenly I'm not thinking about gem publishing or macOS APIs. I'm thinking about protocols and data flows and why Claude can’t get the walkthrough tooltips to line up right.
Task-Oriented Windows. Inside each session, windows become specialized tools:
Window 0: The command center (usually running Claude Code or whatever AI agent is helping today)
Window 1: Vim (mostly just to read Markdown files at this point)
Window 2: Tests, builds, servers
Window 3: A spare shell, for whatever else I might need
Efficient Context Switching. Ctrl-b s shows all sessions. Ctrl-b w shows all windows. Your fingers learn these patterns like a pianist learns scales, and suddenly you're not thinking about how to switch contexts, you're just thinking about where you want to be.
Want to see it in action rather than read about it? I presented this workflow to Latent Space’s AI In Action Club a few weeks ago. You can check out the video here on YouTube.
The Learning Curve That Ate My Twenties
I don’t know how long it actually takes to stop thinking about tmux commands and just doing them. But I promise, one day (I can’t tell you which day, it happens gradually… and then suddenly) your fingers start moving without your brain’s permission:
It can take forever to get comfortable with this workflow. The key bindings are arcane and the concepts are weird if you're used to GUI applications.
But even with all my hard-earned muscle memory, I hit a wall at around three concurrent projects. Beyond that, it becomes like trying to remember which dream you're in while having multiple dreams simultaneously.
And if I'm struggling with more than three projects what hope does everyone else have?
The Great macOS Workspace Disappointment
We've built all these amazing tools for parallel processing in our computers, but we're still using interfaces designed for serial thinking.
We don't actually have good tools for parallel work.
macOS Spaces? Please. They're like someone designed window management after only hearing rumors about what windows are supposed to do. The APIs are basically non-existent. It's like Apple wants you to only work on one thing at a time
Most people end up in this frantic alt-tab dance:
Command-tab to switch applications
Lose track of which terminal has which project
Accidentally run
git pushin the wrong directorySpend ten minutes figuring out where they were
Give up and work on one thing at a time like a sensible human being
But I don’t want to be forced to be sensible. I want to work on ALL MY PROJECTS AT THE SAME TIME.
So We’re Building Something
This is where APM comes in. Actions Per Minute? Agent Project Manager? Agile Project Management? Absolutely Parallel Madness?
I don’t feel very strongly about it, it’s really up to you to decide. At least we have all the domains, so use whichever makes the most sense to you.
Anyway, APM is a gaming-inspired command center for working with AI agents across multiple projects. Think of it like my tmux process but for people who don’t have 15 years to spend memorizing keyboard shortcuts.
The idea is stolen directly from gaming, because gamers figured out something important: if you're going to manage complex parallel systems, you need a HUD that doesn't make you want to cry. They've been managing resource economies, unit positions, and build queues since StarCraft. Meanwhile, we're over here trying to remember which terminal has npm run dev running.
You can still use tmux if you want (I will), but you could also use Cursor, or VSCode, or Zed, or whatever text editor for whatever programming agent you’d like. The point is to make working on multiple projects actually feel like you're in control.
The Parallel vs. Sequential Question
Which brings me to my question for all of you: what are you doing with all this AI acceleration?
Are you like me, trying to juggle more projects simultaneously? Using the speed gains to increase your concurrency rather than your free time?
Or are you like Justin Searls, who built a Discord bridge and now gives Claude Code commands from his Apple Watch while living his actual life? (Which, let's be honest, is probably the healthier approach.)
I toggle between both modes. During the week, I'm the caffeinated octopus, all tentacles and concurrent sessions. On weekends, I try to single-thread: pop in to my office every once in a while and move Claude a step forward while I remember what sunlight feels like.
A Call from the Terminal
To get more specific:
Are you working with multiple agents in parallel? What's your setup?
I want to know:
How many projects do you juggle at once?
What tools are you cobbling together to make it work?
Are you team Multiply-the-Projects or team Live-More-Life?
Have you built any weird contraptions for managing AI agent conversations across different contexts?
Are you doing something completely different that I haven't even thought of?
I suspect we're all making this up as we go along. Nobody Knows How To Build With AI Yet. AI agents are new. Parallel development workflows are still weird and niche. The tooling is... well, let's just say there's room for improvement.
But maybe that's okay. Maybe the best solutions come from a bunch of programmers sharing their weird setups and learning from each other's struggles.
I'd love to hear your stories, especially if they involve Apple Watches, Discord bridges, or other wonderful complications.
Drop a comment, send a message, write a blog post of your own in response! I want to know how everyone else is handling this new world of AI-accelerated development. Are we all just dealing with different flavors of the same chaos?