# SIG-P4B 2026-07-13 — Session Notes

Participants:
  - rafa (UTC+1)
  - Artem | UTC+1
  - Timber | UTC-7
  - NicolasCero
  - Anuraj R | UTC+0300
  - PAtwater | UTC-8 | 31P

---

## 📖 The Reading
Not clearly identified in the discussion. This session was primarily a live tool demo rather than a discussion of a shared text. A prior guest presentation by "Bell/Belle" is referenced but not detailed in the transcript.

## 🧭 Overview
Most of the session opened with logistical trouble getting a guest (Bell/Belle) into the Discord call, which prompted a broader complaint about Discord's friction for outside participants. The bulk of the meeting was a live demo by Artem of **Harmonica**, an AI-agent-mediated tool for collective sense-making and facilitation. The group discussed how Harmonica works, saw a live session created and a Polis-style clustering visualization, and explored whether the group could use it in an upcoming session.

## 💡 Key Points & Themes
- **Discord as a barrier to guests (rafa, Timber):** Bell/Belle struggled to join because Discord requires account creation and captchas. rafa concluded that for external participants they may need to use Google Meet or Zoom instead; he ultimately sent Bell a Google Meet link. Timber noted this has consistently been a "drag" for guests and "drive-by" participants (mentioning that "Bond" hates Discord despite being a good participant).
- **Artem's background:** Artem has followed Summer of Protocols and rafa's writing (since ~2021, while writing a master's thesis on decentralized governance). He worked as a governance lead in a small DAO (Protein Community) and was a member of "Folklore." His practical struggles with sense-making and community engagement motivated building Harmonica.
- **Core premise of Harmonica:** Common collective sense-making methods (retrospectives, Delphi method, road-mapping, business model canvas, value proposition canvas) are essentially *protocols*. Harmonica executes these protocols using a language model to help a group benefit from them.
- **Agent-mediated, MCP-integrated design (Artem):** Artem demoed driving Harmonica from Claude via a custom "Harmonica chat" skill connecting to an MCP server (under the hood, an API). It's being built to be usable by agents, not just as a web app. Given a topic, the agent selects from a seeded template library (it suggested a "proposal forming" template for the demo topic).
- **How a participant session works:** The agent generates a public shareable session URL. Participants enter a ChatGPT-like chat where the AI drives/prompts the conversation, asks some deterministic pre-survey questions, and cross-pollinates by asking participants to react to others' contributions. Responses feed into a "session memory/knowledge" that can produce a visual artifact or be exported as JSON.
- **Asynchronous, scalable facilitation:** Artem framed Harmonica as solving the difficulty of scaling structured conversation beyond ~8–10 people. rafa described it as an agent creating an "asynchronous focus group" and then synthesizing results for participants.
- **Polis-style broad listening (Artem):** Artem showed a visualization (~70 votes on statements) resembling Polis: participants agree/disagree with statements, results cluster, and you can then deliberate on an individual cluster. Statements are "anchored"/emergent — seed statements come from initial sessions, then ~10x more people vote, then clusters can be deliberated on. He cited Audrey Tang's term "broad listening."
- **Iterative loop as the key value (Artem):** The strength is an emergent, open, iterative cycle — seed statements → voting → clustering → deliberation on artifacts. He positioned Harmonica between the *scalability* of a Google Form and the *depth* of a facilitated workshop, letting people reflect on prior rounds and see others' responses.
- **Relevance to the group's own work (rafa):** rafa noted the group is currently doing async work in Google Docs + LLMs (mentioning Venkat working on "the AISA" with rafa and Sachin), effectively a "proto form" of agent-mediated knowledge-product development, and that Harmonica seems aligned with what they're trying to do.

## 🔀 Questions & Disagreements
- **rafa's clarifying question:** Is Harmonica *running* the brainstorm, or *walking participants through* participating in it? Artem clarified it is an agent-mediated process where the agent sits between participants, driving the conversation and synthesizing.
- **PAtwater's questions:** Is the link live/shareable so water colleagues could answer asynchronously (like a focus group)? And what does the output look like? Artem confirmed it works asynchronously; output details were only lightly addressed in the transcript.
- **rafa's open reservation:** He wondered aloud whether the technology is "mature enough" yet for the complex async discussion (AISA) they're attempting.
- No substantive disagreements were noted; the tone was exploratory and supportive.

## 🔗 References Mentioned
- **Harmonica** — Artem's AI-facilitation tool (the subject of the demo).
- **Polis (POLIS)** — referenced as the model for the voting/clustering visualization.
- **Audrey Tang** — credited for the term "broad listening."
- **MCP (Model Context Protocol) servers** and **Claude** — the technical stack used to drive Harmonica from an agent.
- **Summer of Protocols** — research program Artem has followed.
- **Protein Community (DAO)** and **Folklore** — communities Artem was part of.
- Sense-making methods: retrospectives, Delphi method, road-mapping, business model canvas, value proposition canvas.
- Tools contrasted: Google Forms, Miro boards, Google Docs.
- **ISO / SOC 2 / standards bodies** — referenced re: an upcoming session on how standards change with probabilistic (AI) software (rafa couldn't recall the exact organization name).
- People named: Bell/Belle (prior guest), Venkat, Sachin, "Bond."

## ✅ Action Items & Next Time
- **rafa & Artem:** Coordinate offline about using Harmonica in one of the group's sessions over the next couple of months, to do live sense-making, test the product, and give Artem feedback. A candidate topic floated: which people/guests the group should interview or invite to discuss protocols.
- **Artem:** Offered to help set up a session, customize the product for specific needs, and contribute to the group's research.
- **Upcoming sessions (rafa):** Attempting to schedule (1) folks from "titles" doing generative art / niche artist models related to marketing, and (2) a standards-body discussion on how ISO/SOC 2 compliance is changing with probabilistic software.
- **Reminder (rafa):** Everyone should review the "KitKraft" workshop, give feedback on desired changes, or sign up to attend — link posted in the "protocols for business" channel.

## ⭐ Memorable Quotes
- **Artem:** "All of those are basically protocols... the tool that I'm going to demo is basically something that can execute those protocols using a language model."
- **Artem:** "I'm trying to find this sweet spot between scalability of a Google Form and also the depth of a facilitated workshop."
- **rafa:** "We're in the proto form of kind of this agent-mediated, like, knowledge product development."

---

*Note: The transcript is heavily interleaved due to overlapping audio and the concurrent Discord login troubleshooting, so some exchanges (especially early logistics and PAtwater's closing questions) are fragmentary. Names like "Bell/Belle" and organizations (the standards institute) were not clearly resolved in the audio.*