# SIG-P4B 2026-09-17 — Session Notes

Participants:
  - rafa (UTC+1)
  - Sachin

---

## 📖 The Reading
Not clearly identified in the discussion. This session was not a reading discussion but a working/planning conversation between rafa and Sachin about the design of a multi-session AI-collaboration workshop and its accompanying slide deck.

## 🧭 Overview
rafa walked Sachin through the current state and flow of a four-session workshop built around AI-assisted collaboration, "kits," and "factories." He explained the conceptual arc of each session, flagged parts of the slide deck (and a "two-by-two") that still need restructuring, and demonstrated how the workshop repo captures append-only logs of participants' and agents' work. Sachin gave feedback (notably to constrain the "factory" exercise) and reflected on how the workshop's persistence in a shared code base makes it different from a typical workshop. The call ended abruptly as rafa had to jump to another meeting.

## 💡 Key Points & Themes
- **rafa — Emergent collaboration via public practice:** Collaboration in AI is emerging from moving *private practice to public practice*, not from intentional workflow design (contrasted with agile/waterfall). rafa suggested this is "the collaboration paradigm."
- **rafa — Stigmergy as the dominant collaboration mode:** Rather than handing off work, you broadcast the work itself — signals about the work, your logs, the production system, the infrastructure you use. He proposed illustrating this with images of a personal journal vs. a field notebook, using ethnographers/bird watchers who "publish their field notes with no intention other than publishing." Sachin suggested plugging in the Humboldt "nature of a mountain" diagram as an example.
- **Session 1 (customization → traces):** Exercise flow — ask your own AI to surface the tools you're using (some you're unaware of), then examine others' inventories to find "near or overlapping" tools. Finding overlapping traces makes the collaboration signal stronger. Ends with joint troubleshooting / office hours.
- **Session 2 (running a kit without the author):** Central question — "Can you run a kit without the author's oversight?" Moving a bespoke, on-your-machine tool toward something where you become interchangeable. Exercises: articulate what still requires *you*, then build hard protocols/standards to translate it — i.e., "automate yourself out of the role of the kit."
- **rafa — Electrification reframing:** Rather than focusing on the reorganization of the production line, the more interesting lens is how *humans themselves were transformed* by electricity into "different types of humans," which then allowed factories to reorganize. Acknowledged this deeper insight isn't strictly needed for the workshop.
- **Factories:** A "factory" makes humans interchangeable in the production process; it need not be a single scale/file — it can be an app, a website, etc. Examples cited: a "brand kit" and a "manuscript factory" rafa put together (debatably "true" factories). Exercise: use your AI agent to help "factorio" (industrialize) a specific kit.
- **Session on bridges:** Once factories work alone, they become networked. The exercise combines two kits (or factory + kit) to produce an output — composing a network of kits together. Framed as the inverse of the earlier "remove yourself from the kit" exercise.
- **Show and tell:** Sessions close with participants presenting a kit, a factory, or a bridge.
- **Persistent, folded memory:** Both rafa and Sachin emphasized that unlike typical workshops (notes get shelved and forgotten), this one persists in the participants' code bases. Participants end up sharing "a slice of each other's memory," the institute retains an output, and future workshops fork the current workshop stack into a new repo to iterate on feedback.
- **Append-only logs:** Copying the template into a participant's inventory/factory generates a log; agents' participation also adds log lines. Participant feedback to their "harness" gets logged and integrated into the ongoing work (the harness has instructions to fill in a log).

## 🔀 Questions & Disagreements
- **Fitting it into four sessions:** Most feedback rafa has received is skepticism that this can fit in four days/sessions. rafa's response: we're "miscalibrating what it means to run a workshop" when assistants automate exercise execution. Sachin agreed there's "no real limit — it just depends on how much compute people are willing to spend."
- **The two-by-two is broken:** rafa flagged that a two-by-two diagram is currently "wrong" and needs fixing before a downstream reflection ritual (noticing precon-modification patterns / external traces / overlaps) can land. He was unsure where "content precon modification" belongs in the flow — an open structural question.
- **Sachin — constrain the factory exercise:** Suggested the factory portion needs more constrained instructions (referenced a 60–90 [minute?] time box). rafa proposed dropping the "build a solution" component so the exercise is just to *identify a kit and create a road map to a factory* rather than actually building one. Two reconceptualization framings offered (using a job-search kit example): (1) something that produces RAG agents for job searching, or (2) making the industry/person using the RAG "robot" swappable.
- **Loose ends:** Whether Sachin's commit had landed in the repo ("Did mine make it there?" — not yet found), and whether session times were updated (rafa unsure but had committed to current times). Both left unresolved as the call ended.

## 🔗 References Mentioned
- Humboldt "nature of a mountain" diagram (suggested by Sachin as an illustrative example)
- Stigmergy (concept — signaling/broadcasting work rather than handing it off)
- Agile / waterfall (as contrasting, intentional workflow-design paradigms)
- C-3PO ("c three PO") — referenced both as an example of emerging interfaces and as a system into which workshop output gets folded/synthesized
- "Factorio" (used as a verb, referencing the game, to mean industrializing a kit)
- RAG agents (retrieval-augmented generation) — job-search example
- Electrification of factories (historical analogy)

## ✅ Action Items & Next Time
- **rafa:** Update slide 10 ("what happens after kits") for clearer communication.
- **rafa:** Fix the two-by-two diagram before the reflection ritual can be finalized; resolve where "content precon modification" fits in the flow.
- **Consider (rafa/Sachin):** Simplify the factory exercise to "identify a kit + road-map to a factory" rather than building one; add more constrained instructions / a time box.
- **rafa:** Locate Sachin's commit in the repo (not yet found).
- **Open:** Confirm/finalize session times (rafa committed to current times but was unsure whether updated).
- Slides are uploaded to the repo under "slides"; Sachin to review.
- Next: the workshop is structured as four sessions (kits → running a kit without the author → factories → bridges), each ending in show-and-tell.

## ⭐ Memorable Quotes
- **rafa:** "You're not actually handing off work. You're actually broadcasting the work itself, signs about the work, your log and the production system."
- **Sachin:** "Usually you do a workshop, you go away, maybe you take some notes, you forget. But this potentially remains in your code base."
- **rafa:** "It's the right level of weird."