--- title: More coding agents don’t need more terminal panes description: After trying cmux, Herdr, and Otty, I realized I didn’t want better splits. I wanted a calmer way to manage coding-agent sessions. type: blog date: 2026-08-14 authors: - name: Little Pan url: https://x.com/heylittlepan cover: light: /blog/agent-waiting-room-light.webp dark: /blog/agent-waiting-room-dark.webp seo: title: "More Coding Agents Don’t Need More Terminal Panes" description: What cmux, Herdr, and Otty taught me about managing coding agents—and why I started building a native Mac terminal mux without split panes. image: /og/more-coding-agents-fewer-terminal-panes.png x: creator: "@heylittlepan" --- *By Little Pan · August 14, 2026* At some point, my terminal stopped being a place where I ran commands. It became a waiting room full of coding agents. *Illustration: several coding-agent sessions wait quietly in different states while one terminal remains in focus.* I’ve been trying a bunch of the new terminal muxes lately. [cmux](https://cmux.com/), [Herdr](https://herdr.dev/), [Otty](https://otty.sh/). They’re all interesting and, annoyingly, all pretty good at what they’re trying to do. None of them stuck for me. For a while I thought this just meant I was picky. Which is probably true. But after bouncing between them, I started to think I might be looking for the wrong thing. I didn’t need a better way to split terminals. I wanted a calmer way to manage agent sessions. So I started building [PonyMux](/). This isn’t really a review of those products. They change fast, and I’m sure there are things I missed or used differently from the people who love them. This is more about what I learned from trying to make them fit the way I work. ## It started with too many agents My terminal used to be mostly a place where I typed commands. Coding agents turned more and more of it into a place where I start something, switch away, and come back later. Once enough of those sessions are running, the problem changes. It becomes less about entering commands and more about knowing which session needs attention, what each one was for, and how to get back to it. The obvious answer is a mux. Put more terminals in one place. Keep them visible. Split the screen until everything fits. This works pretty well for ordinary terminal jobs. A server on the left, logs below, the shell you’re using on the right. Agents are a little different. They’re long-lived, asynchronous, and full of context. Most of the time I don’t need to watch one work. Watching text scroll past is not the same as knowing what needs my attention. Still, I assumed the answer was a better mux. So I went looking. ## cmux: a better command center [cmux](https://cmux.com/) is native and fast, and its vertical sidebar makes sense for agent work. It adds status indicators, split panes, an embedded browser, and programmability around the terminal. It’s a coherent answer to the problem: make more work visible and controllable in one workspace. What it didn’t change for me was the spatial part of the job. More panes still mean deciding what should stay visible, how much room it gets, and how the workspace should be arranged. That’s exactly the work I was hoping a new kind of mux would remove. That’s not a cmux problem. Plenty of people want a command center. I was slowly realizing that I don’t. {/* Draft note: replace the generic layout paragraph with one exact moment from using cmux: what was open, how many panes, and what made you give up. */} ## Herdr: the mux is the product [Herdr](https://herdr.dev/) goes in a very different direction. It lives inside the terminal you already use. Sessions keep running when you detach, they can live on remote machines, and Herdr knows whether agents are working, blocked, or done. It’s a clean idea. No account, no big desktop environment, no pretending that the terminal isn’t still a terminal. But I spend most of my time on a Mac, and I realized I wanted the opposite boundary. I wanted a native app to be the home for all these sessions. I wanted finding and organizing them to feel more like navigating Things or Bear, and less like operating another interface inside my terminal. Herdr made the multiplexer much better at understanding agents. I was beginning to wonder whether I wanted to feel the multiplexer at all. {/* Draft note: add the real Herdr workflow that made this mismatch clear. Keep it as “not for me,” rather than a universal criticism. */} ## Otty: how much should the terminal own? [Otty](https://otty.sh/) is native, polished, and much more ambitious than a traditional terminal. It has agent state, history, recovery, prompt workflows, panes, previews, and a lot of thought put into making agents feel at home. It sits somewhere between a terminal and a full agent development environment. That’s a reasonable place to build a product. It’s just not where I wanted to end up. Otty’s direction raises an interesting question: how much of the agent workflow should the terminal own? My answer kept drifting toward “less.” I didn’t want the terminal that could contain the most workflow. I wanted the smallest mental model that could support a lot of work. {/* Draft note: name the one or two Otty features that felt like the boundary between a terminal and an ADE for you. */} **Three ways to frame the terminal** - cmux — command center - Herdr — agent-aware mux - Otty — agent workspace Three muxes, three different ideas about what belongs inside the terminal.

They weren’t failing at the same job. They were solving the jobs I thought I wanted.

## I had the problem backwards I kept comparing features. Persistence, notifications, panes, agent state, remote access, browsers. That was probably the wrong way to look at it.

The feature I was looking for was subtraction.

The thing I was running out of wasn’t screen space. It was attention. Most of the time I only needed to know three things: - Which agent needs me now? - What was this session about? - Can I come back later without reconstructing everything? I didn’t need every agent to remain visible. In fact, keeping them visible made them compete for attention even when nothing interesting was happening. With enough panes, I wasn’t coordinating more work. I was building a dashboard I had to babysit.

I didn’t want a better pane manager. I wanted to stop managing panes.

## So: no splits The first idea behind PonyMux was almost suspiciously simple: one terminal session in a window, with enough room to breathe. Everything else could live outside that surface. A list for scanning sessions. Folders for grouping them. Facets for narrowing a large pile down to what matters right now. A Quick Terminal for the small command that doesn’t deserve a permanent home. And no split panes. **Same sessions, different mental model** - Keep everything visible — manage space. - Keep one thing in focus — manage attention. More sessions don’t have to mean more terminals on screen. Admittedly, “a terminal mux without splits” is a slightly strange product pitch. Splits are treated as table stakes. Removing them looks less like a design decision and more like I forgot to finish the feature. But that constraint is the whole point. PonyMux should help me run more sessions without asking me to look at more terminals. There’s another part to it too. An agent terminal no longer feels disposable. It has a project, a purpose, a history, and a thread of thought I may want again next week. It should be something I can recognize and recall, not just a rectangle that happens to be open. Every terminal should have a memory. Not “let’s put another AI inside the terminal” memory. Just enough identity and context that I can find the right session and understand why it exists. ## Maybe this is just me Some people genuinely think spatially. They want four panes visible, a browser beside them, logs in the corner, and the whole system under their eyes. cmux, Herdr, and Otty each serve versions of that workflow better than PonyMux is likely to. There’s an obvious answer that already looks closer to what I’m describing: the Codex app and the Claude app. Neither one asks you to keep a wall of terminal panes visible. Each task has somewhere to live, and you move between them when you need to. In that sense, they’re pretty good evidence that this way of working isn’t just in my head. I like that direction. I just don’t always want the rest of the app around it. Both do a lot more than a terminal does. That can be useful. It also means bringing a much bigger app into a moment when I may only want a terminal.

What I keep wanting is a smaller version of the same idea: the focus of an agent app, with the lightness of a terminal.

Maybe that is a narrow preference. PonyMux isn’t trying to replace those apps, or win a feature checklist. It’s for people whose terminal has filled up with long-lived agent sessions, but who don’t want their attention to fragment along with it. People who would rather switch deliberately than monitor continuously. People who want this part of the day to feel a little calmer. I don’t know yet how many of those people exist. I know I’m one of them. I started building PonyMux with Claude Code and Codex CLI, which is maybe fitting: agents helping build a terminal for managing agents. Slightly circular, but here we are. PonyMux is still early. But I know what I want from it: one terminal in front of me, the rest easy to find, and no layout to babysit. More as I build.