Ettle

ettle

Everyone on my team works through Claude Code all day. We message each other when something has already gone wrong. There's no standup to skip and no meeting to fix — the work just runs into itself, and you find out as a merge conflict, or as an afternoon two people spent building the same thing.

That gap used to be covered by the implementation window. Building was slow enough that alignment happened in the seams: a message while someone was still writing the code, a comment on a draft PR, a question in a meeting that was going to happen anyway. Agents collapsed the window, the seams went with it, and the pull request now carries every checkpoint that used to be spread across a week — arriving at the one moment when changing course costs the most.

ettle puts the checkpoints back without adding a place to go. Each person's agent distills their own session into typed atoms — what they're working on, what they've committed to, what they depend on, what they're assuming. Only the atoms cross; the raw session never leaves the machine. A reconcile pass compares them across the team and surfaces just the friction — the dependency you're about to break, the work two of you are duplicating, the assumption you're holding that someone else quietly dropped — privately, in your own next session, before it's the thing you'd have called a meeting about.

The other answer to this is a shared workspace: move the team into one multiplayer room and coordinate there. (GitHub Next's Ace is the clearest version — [One Developer, Two Dozen Agents, Zero Alignment](https://maggieappleton.com/zero-alignment) is worth your time either way.) That works if your team will move. Mine won't: we already have the shared room, and we route around it. So ettle assumes the opposite — nobody changes tools, nothing lands anywhere shared unless you decide to put it there, and it's useful at N=1 before a second person ever joins.

⚠️ **Very early — a design-stage proof-of-concept, published w