Community scene

Work out how Commonplace can help build an intellectual community around deep questions about the new computing model: harness design, context engineering, codification and the knowledge systems agents work in. The operator stated this as a project goal on 2026-10-24, after the Borretti "No Man Is an Island" ingest, and commissioned this workshop the same day. The operator does not yet know how to start; the workshop exists to find a first concrete step, not to plan a programme.

The operator's premise: discussion of the new computing model can be as deep as the old discussions of programming languages and libraries. Prompting is part of it but not the whole, and the visible discourse is mostly marketing. That premise is a hypothesis; see the Limitations of the Borretti ingest and the depth-versus-discourse entry in kb/log.md.

Constraints

  • Do not present Commonplace as the counterexample to claims that the field is shallow. Outside-facing arguments rest on outside evidence; the operator called self-citation here a faux pas.
  • The goal is recorded only here. Adding it to the AGENTS.md Purpose needs the operator's wording or approval.

Data points

Two cases so far. Two cases do not establish a mechanism.

  1. Borretti's model of why communities fail. Sustained private work needs a shared body of work and extrinsic motivation (others use and credit your contribution). AI absorbs contributions without credit, so the loop breaks. He also treats agent readers as no audience. Read as a list of what a community must supply, not only as a prediction.
  2. Fabryka AI, a Warsaw lab the operator judges to be a working scene. Observed from its home, story and research-notes pages on 2026-10-24 only, not from its Discord or contributor numbers:
  3. a concrete shared target with outside scoreboards ("I thought I could build a model better than Bielik"; leaderboards, benchmarks);
  4. resources individuals lack (compute, experiment tracking, inference API);
  5. AI as the entry point: an agent trained a model, and curiosity moved from "Which model should I use?" to "Why does this one work better?";
  6. credited, inspectable output: notes pinned to commits, negative results published, contributor profiles, and an Application Book where outsiders submit a use case and the lab helps publish it;
  7. a bounded identity (Polish, small models) that large labs do not serve.

Gap against Commonplace: methodology quality has no outside scoreboard, so there is no obvious counterpart to "beat Bielik".

Candidate first step

An Application-Book analogue: an outsider brings an agent or memory system and receives a published, credited analysis they can dispute, produced with analyse-agentic-system and comparable with the existing agentic-system reviews. It would supply Borretti's two inputs (credit and a shared, comparable body of work) and treat AI tooling as the entry point. Untested; weaker alternatives are an open problems list and working papers, which add shared work but no first step for a newcomer.

Open questions

  • Who is the first audience, and where do they already gather?
  • What could serve as a shared target or scoreboard without rewarding easily checked proxies over methodology quality?
  • How do outside responses come back into the KB (ingests, disputes on working papers, credited contributions)?
  • What role, if any, do agent readers have in the community, given Borretti's opposite premise?
  • More data points: other scenes that formed around a new technical field.
  • volunteer-compute-tasks — contributor tasks with cheap verification; a possible entry route for outside contributors.
  • first-downstream-run — an experiment template for an outside experimenter.
  • dsw-talk — a public practitioner talk; one channel to an audience.

Closing

Close when one first step has been chosen and tried, or the operator drops the goal. Extract what was learned about scene formation to kb/theory/ only if a mechanism emerges beyond these case descriptions; otherwise record outcomes in kb/log.md.