The deliberation loop
The methods consultants trust were built for a room of eight. Run them async, at scale, and they fall apart in the same place — which turns out to be a fixable place.
Blog
Notes on AI-facilitated structured conversation: product updates, case-study deep-dives, and what we're learning from running sessions in the field.
Harmonica 0.5: the project layer
0.4 made sessions chain into methods. 0.5 makes the project a place that accumulates: sessions feed shared knowledge and shared context, and that substrate flows back into the next session.
Harmonica 0.4: a runtime for facilitation
Harmonica began as an AI interviewer, with 1:1 chats in isolated sessions. Version 0.4 turns it into a runtime: something that takes a facilitation method and runs it reliably in multiplayer mode.
We shipped prompt improvements against a broken scoreboard
We built automated quality scoring for Harmonica's AI facilitator, then discovered the scoring itself was wrong in three distinct ways. The fixes took longer than the original prompt improvements — and mattered more.
Retrospectives shouldn't be projects
Anthropic just shipped an AI feature called Dreams. Organizations are missing the same rhythm — and paying six figures every few years for it.