## Who owns what in AI, and the ABP in the interview

Two asks from the lead in one note ([D20](briefs.html)).

- **[A new article](article-who-owns-what-in-ai.html)**, the first of a kind the lead wants to
  write more of: take something strong seen on LinkedIn and put our world on top of it. Here it is
  an infographic, *Who owns what in AI (By Role)*, by Alex Miguel Meyer, which maps thirteen roles
  to what each owns in AI and what that protects. Its rows are quoted and credited; its image is not
  reproduced.
- **The argument is that the map runs sideways and accountability runs upwards.** The map's middle
  column says where a risk lands. The organisation chart says where the risk goes when that holder
  cannot decide, or does not. Every company redraws both, and the article says which parts may
  change and which stay fixed.
- **One invented agent is worked from the eight rows of its behaviour policy to the board.** A sales
  assistant connected to a rep's mail, calendar and CRM gives five risks. The article sets out:
  - the same exposure in thirteen voices, and one risk followed up the chain;
  - seven rules, and six weeks of the loop;
  - the map again, with four more columns: what each role holds, what it may accept, who it
    answers to, and whether it can stop the agent.
- **It is written to be built from.** Section 09 is the model: the node types, edges that each have
  an inverse, and the computations for placement, the clock, ceasing, the stop check and the board's
  view. Two rules are new and say so: authority handed down from the board's risk appetite as kinds
  of consequence and lengths of time, and placing a risk by the map. Nothing is scored.
- **[The founder interview](interview-founder-marketing.html) now asks about the ABP itself**, in a
  first part of six questions:
  - its name;
  - whether the interviewee can explain it back after hearing it once;
  - what it adds;
  - whether the market understands the problem;
  - its value to the people who would use it;
  - whether it should sell, and at what price.

  The description it works from is the site's own. The interview is now thirty minutes, with a
  sixteen-section summary. The copy was tested in a browser and is exact.

What was not done:
- The LinkedIn post's own address and date are not on the page; the article says when the
  infographic reached us, and the link is the lead's to add.
- The longer prompt has not been run in ChatGPT voice mode.
- The seven roles the article needs and our model lacks are not added: COO, CIO / Data, Legal, HR,
  Marketing, Sales and all employees.
- Authority as data, and an engine that runs the clocks, are proposed and not built. The article's
  section 10 says so.