# SIG-P4B 2026-09-22 — Session Notes

Participants:
  - rafa (UTC+1)
  - Sachin
  - doxxed_as_zak
  - Eric Alston (UTC - 6)
  - PAtwater | UTC-8 | 31P

---

## 📖 The Reading
Not clearly identified in the discussion. This session was a working/demo session of a "KitKraft" workshop (part of a recurring special interest group) rather than a discussion of a specific text. References were made to prior sessions, a "water data" project presented by Maxwell, and a "psycho history" special interest group that reads history books.

## 🧭 Overview
This session was part of an AI/vibe-coding workshop in which participants demonstrated and discussed self-built software tools. rafa demonstrated a distributed robot-hiring/contracting project and a teleoperated robot, then rafa and Sachin worked through their emerging vocabulary for what they call "factories" (personal, protocolized AI-enabled software that isn't quite a product). The conversation broadened into a recurring theme raised by doxxed_as_zak — the lack of shared language across tools and projects (a "Tower of Babel" problem) — before turning to logistics for an afternoon show-and-tell session and a closing exchange with Eric Alston on how "writing for an external audience" improves both human and AI output.

## 💡 Key Points & Themes

**"Factories" as a new category of software (rafa, Sachin)**
- rafa demonstrated Yak Robot (yakrobot.com): distributed robot hiring/contracting on blockchain rails. A person submits an RFP/task in English, it's converted to machine-readable tasks, robots review against their capabilities, bid, win, get paid, execute, and return data. rafa said this "works ish" on mainnet.
- rafa is also building an LLM data pipeline that converts English PDFs into machine-readable robot tasks (e.g. "take the beer out of the fridge" → specific location/action).
- rafa distinguishes a **"factory"** from a product/platform: it's a heavily protocolized process built for oneself or friends, not for commercial scale. Framing something as a "product" pulls you into a different design paradigm.
- Sachin: a factory is essentially "an extension of a single person" — embodied skills turned into software that can run without them, plus affordances like automation, densification, and bespoke creation (concepts from the prior day).
- rafa: these are objects that "shouldn't exist" under the old paradigm — previously not cost-effective in time, money, or skill, now possible. They're more than a toy, hardened enough to build a life/community on, but not at scale and not quite open-source software.
- doxxed_as_zak likened this to the 1980s home-electronics/component-store movement — hobbyist building that later disappeared as products matured.
- rafa's and Sachin's bet: a few decades of cost-effective bespoke "micro-factory" building, after which a new class of products emerges and future generations tinker less.
- Sachin: what's easiest to turn into a factory is a domain where you already have deep context (years of publishing experience, a 5-year robotics project). Model companies tend to productize high-context domains (e.g. OpenAI releasing a finance feature in ChatGPT after many people built personal finance apps via vibe coding).

**The "Tower of Babel" / shared-language problem (doxxed_as_zak, rafa, Sachin)**
- doxxed_as_zak raised (repeatedly, across overlapping speech) that each personal tool/harness develops its own domain language, so agents and people can't easily understand or integrate each other's projects. He proposed the need for a "parent domain language" for software, and mused about a "design/art-history education for software" to build shared aesthetic vocabulary. He framed software as uniquely prone to "semantic mess" ("let x equal anything").
- Sachin experienced this concretely: he found it hard to gain context on Patrick's kit and had to ask Claude to "translate" it for him.
- rafa: there's an "uncanny valley" — a tool customized for you fails on both counts: not customized enough to hand off usefully, yet not general enough to be a product. He's used their group's "lens" to have a friend explain a skill he couldn't understand.
- rafa's anecdote: a team member built a great kit no one could adopt (weird language, over-customized), and no one told them.
- rafa predicts a "complete mess / Tower of Babel" over the next ~10 years, then convergence.

**How emerging fields solve the language problem (rafa)**
- rafa: crypto solved this via (1) standardized language for protocol upgrades and (2) the conference circuit as a consensus mechanism for terminology (plus Twitter). He expects similar convergence, new standards bodies, and new verticals.
- rafa emphasized IRL / peer-to-peer co-building (like this workshop) as increasingly important in the pre-product phase, invoking a "Gramsci gap" — a time of monsters between paradigms — after which the new platform's ontology dominates and history gets rewritten as if the products always existed (analogy to forgotten 1990s internet protocol chaos).
- Sachin tied this to Maxwell's "water data" project: different water bodies used different measurement standards, making water rates incomprehensible; the AI solution was translation across existing standards rather than inventing yet another standard.

**Writing / externalizing thought improves output (Eric Alston, rafa)**
- rafa's practical tip: tell Claude to "clean this up, remove personal info, add a README/orientation, and prepare it so a friend can use it, then upload to GitHub." Forcing the AI to produce something that works without your local history/context/tool-chain surfaces bugs and generalizes the tool.
- rafa observed the AI enters a "different mode of creation" and produces better output when told to prepare something to teach/hand to someone else — a lesson he learned only by actually trying to send output to a friend and finding it broken for anyone but himself.
- Eric Alston drew a strong analogy to writing and academic conferences: externalizing thought for an audience "imposes order" cognition otherwise lacks — even journaling is writing for a future reader.
- Eric also described intentionally telling one model that its draft would be reviewed by another (e.g. ChatGPT's top tier), which he feels yields better outputs — while noting he might be over-anthropomorphizing.
- rafa: analogous to hosting a weekly dinner (makes you cook/clean/tell a cleaner story) or the mental shift of the Substack "publish" button. He and Sachin noted the difficulty of explaining concepts they've internalized over ~1.5 years to newcomers.

**Other demos / projects mentioned**
- rafa & "Geraldine" (Geraldine Andres?) prototyped an MVP to help immigrants in Germany with job search — matching qualifications and clarifying actual language requirements (many German listings don't truly require German), with questionnaires (shifts, relocation, license) and suggestions (e.g. move from A2 to B2, certify a license), delivered in the user's language. Motivated by the inaccessibility of German employment offices to non-German speakers.
- rafa mentioned "The Prime Radiant" from the psycho history special interest group — studying "world machines," a "modernity machine" that created standard practices, a current "divergent machine" phase, an "almanac" forecasting future technological paradigms, and an oracle for asking "what if" historical prophecies (e.g. a 5-year global oil shortage). rafa framed this as only possible now that AI can synthesize history at scale, and stressed that imagination — not software-building cost — is the scarce resource.

## 🔀 Questions & Disagreements
- **doxxed_as_zak's open question:** "Will we apply domain-driven design to language itself?" — asking whether/how a shared or "parent" domain language for software can emerge. rafa flagged it as important and unsolved, betting on a decade of mess before convergence. No resolution.
- doxxed_as_zak initially questioned rafa's use of "factory" for what looked like just remote-control software; rafa and Sachin clarified the distinction (protocolized personal process vs. commercial product). This was clarification more than disagreement.
- Eric Alston noted uncertainty about whether telling models they'll be "reviewed" by another model genuinely helps or whether he's "overly anthropomorphizing."
- Logistical confusion: Eric asked when the show-and-tell session was; clarified as 3–4 PM Central / 2 PM Mountain, ~3.5 hours later. (Eric can attend only the first half — kindergarten pickup.)

## 🔗 References Mentioned
- Yak Robot — yakrobot.com and register.yakrobot.com (rafa's distributed robotics project)
- "The Prime Radiant" / psycho history special interest group — studying "world machines," "modernity/divergent machines," an "almanac," and an oracle
- The "water data" project presented by Maxwell (standards translation across water bodies)
- Protocol Brand Kit (mentioned by rafa as an analogy for non-commercial tools)
- Claude / Claude Code; ChatGPT / OpenAI (models and tools discussed)
- GitHub (shared kit repos)
- Substack (the "publish button" analogy)
- Crypto conference circuit and Twitter (as language/standards consensus mechanisms)
- Antonio Gramsci — the "Gramsci gap"/"time of monsters" concept (invoked by rafa)
- Internet protocol history — TCP/IP, HTTP/HTTPS (rafa struggled to recall exact acronyms, self-noted)
- "Geraldine" (Geraldine Andres) and "Andres" — collaborators on the Germany job-search MVP (appear to be workshop participants not on the roster)
- Mike, Andre/Andres — mentioned as having built things to demo (third parties)

## ✅ Action Items & Next Time
- **Closing show-and-tell session** scheduled 3–4 PM Central (2 PM Mountain), same day, for participants to present what they hacked on.
- **Eric Alston** agreed to present (first half of the slot only) — a "banal" file-organization kit using Claude Code to clean up years of downloads/duplicate drafts across ~16 external hard drives before migrating to a new MacBook.
- **PAtwater** offered to show something "janky" for his RAG factory.
- **Sachin** agreed to demo and was asked by rafa to also present the "Jamverse" work.
- rafa hoped Mike and Andre/Andres would demo their projects.
- **Suggested workflow (rafa):** as you finish a one-shot kit, prompt Claude to clean it up, strip personal info, add a README/orientation, and push to your GitHub folder so others can inspect and reuse it.
- A planned "kit swap" (fork and improve someone else's kit) was **deferred to a future run** of the workshop due to time.
- rafa and Sachin offered to stay on / be pinged to "rubber-duck" and troubleshoot participants' kits.

## ⭐ Memorable Quotes
- **rafa:** "It's software that, in theory, shouldn't exist... These are all things that would just not have gotten created, but they're also not at scale. They're like home brew, but... you will be able to build your life as a community, as an individual on top of it."
- **rafa:** "There's a real Tower of Babel situation going on."
- **Eric Alston:** "It's this externalization that imposes order in ways that ordinary cognition simply doesn't tend to."