2026-08-04 Tue
Responding to yegge; considering terminal multiplexers
- Person/Steve Yegge posted a pair of essays: The Shape of Things to Come, Part 1: The Continuous Thunderdome and Part 2: Model Welfare for Agentic Engineers. Saved both to Readwise/Reader; haven't read them yet.
- The bit from Part 1 that interested me is that the whole system started 6-7 weeks ago with him asking Claude for an Emacs interface to switch between agents, and everything after that grew organically out of asking for more features.
- That reminds me of herdr, a Rust tmux alternative aimed at organizing parallel CLI coding agents. Been meaning to try it; haven't yet. Weighing it on Terminal/Multiplexer/Idea against Zellij.
- I've glanced more than once at Person/Chris Power aka
typecraft's Person/Chris Power/GitHub/dotfiles and learned a bit about nvim and tmux from them. He recently released typecraft - ditching tmux for herdr. - Claude/Desktop/Code and the Codex code tab do seem well-incentivized to get more agent-oriented over time. Some nice things in them: commenting directly on parts of a plan in Claude Desktop, spinning off a worktree, a heads-up display of which agents need replies.
- That said, I'm dissatisfied with how they don't surface the right info in the right places, and don't make me feel like they're helping me organize the information buried in the chat threads. Considering herdr and other interfaces as a way to get a better handle on actionable items.
- What drives me nuts in particular:
- The auto-suggested thread names are terrible in basically every AI provider.
- Claude doesn't let me group by directory the way Codex does.
- On-the-fly clusters and higher-level planning constructs are seemingly completely out of scope for these interfaces.
- I spent significant time with Beads and ended up abandoning it — it became total slopware, technically complicated and basically unusable. That said, I like how Yegge rethinks things. Hopefully more organizations attempt an original path like he does.
Filed