SIG-P4B 2026-09-17

Session Recording Summary · 19m 31s · Full notes ↗

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.

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.

• **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.

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 tw

Participants: rafa (UTC+1), Sachin