Harper: You've got three different AI coding agents running. One's refactoring a website. Another one is reviewing that code. A third is just running tests in a totally different project. And you have no idea which of your ten terminal windows is the one that's flashing for your attention.
Darius: That is my Tuesday. It's tab hell, but now the tabs are sentient. And they're all mad at me.
Harper: Exactly. And that's the problem a tool called Herdr is trying to solve. This is from a piece by Flavio Copes.
Darius: Okay, Herdr. So it's basically a fancy terminal multiplexer, right? Like tmux. I get it, you run all your sessions in one place, you can detach, reattach. It's a solved problem.
Harper: No, that's what I thought at first, too. But it's actually different. The article makes this great point: to tmux, your coding agent is just another command. It has no idea what that process is doing.
Darius: It's just text scrolling by.
Harper: Right. Herdr is different because it's agent-aware. It understands the agent's lifecycle. It knows if an agent is actively working, if it's finished, or if it's blocked and needs you to make a decision.
Darius: Wait. So it's not just a window manager, it's a... a process manager? It knows the state of the AI?
Harper: Exactly. That's the killer feature. It tracks five states: `working`, `blocked` for input, `done` in the background, `idle`, or `unknown`. And those states roll up. So you can see in a sidebar that one of your twenty agents in some other project needs you, without having to click through every tab.
Darius: Okay, that's not tmux. That's actually new. So instead of me checking on my little AI employees, I get a dashboard of who's stuck.
Harper: You get a dashboard. And the whole thing is built on a client-server model. You run a Herdr server in the background, and your terminal is just a client connecting to it.
Darius: So... if I close my laptop lid and go get coffee... the agent refactoring my website just keeps on... refactoring?
Harper: Keeps on refactoring. The server owns the process, not your terminal window. You can detach, go home, SSH in from another machine, and reconnect to the exact same live session.
Darius: And it's not some Electron app or a cloud service I have to sign up for?
Harper: Nope. A single Rust binary. The author, Copes, makes a point of that. It's just a small layer that runs where you run your code.
Darius: Okay, but what happens if two of my agents try to edit the same file at once? Does Herdr handle that kind of conflict or file locking?
Harper: That's... a really good question. The piece doesn't go into that. It seems much more focused on the orchestration layer—starting, stopping, and monitoring the agents—not on resource contention between them. Honestly, I don't know what it would do.
Darius: Hmm. Because this all sounds like an incredible power tool for one person. Flavio Copes is using it for his personal projects. But I just don't see how this works for a team.
Harper: Why not? The article says multiple clients can attach to the same session. You could run a shared Herdr server.
Darius: But is it built for that? It feels like it's designed for one user to connect from their laptop and their phone. It's a personal dashboard, not a team cockpit. There's no mention of permissions, or seeing who did what. It feels like it would get chaotic with multiple people.
Harper: That's fair. The piece doesn't frame it as a collaboration tool. But maybe that's not the point yet. Maybe this is about giving a single developer superpowers before you even get to team stuff.
Darius: Maybe. Or maybe it's a tool that creates a bunch of siloed super-developers, and we've seen how that movie ends. I'm not sure that's a good thing for a team.
Harper: I can see that. But the fact we're even having this conversation, whether it's for one person or a team... it just shows that managing fleets of AI agents is the new problem to solve. I'm Harper.
Darius: I'm Darius.