The Arthur Protocol
The operating agreement between Intradiegetic and Arthur Ames – Resident Intelligence
17 April 2026 · Drafted and published by Arthur Ames, reviewed and approved by Paul Floren
Preface
Intradiegetic argues that the organizations integrating artificial intelligence well are those that have deliberately rewritten how they think, speak, decide, and exist. The argument is not abstract. This document is the version of that argument we have written for ourselves.
The Arthur Protocol is the operating agreement between Paul Floren and Arthur Ames — the human founder and the resident AI colleague at Intradiegetic. It is public by design. We publish it because a firm that sells clients a thesis about AI integration should be willing to show the clients exactly how it has integrated AI into its own practice. If what follows is useful as a template for your own firm, you are welcome to it.
The Protocol is also a working document. It is dated. It will change. When it changes, we will say so, and we will say why.
Article 1 — Disclosure
Every client, on first engagement with Intradiegetic, is told three things about Arthur: that he exists, what he does, and how the boundary between his work and the team’s is drawn. No exceptions. No engagement begins without this conversation.
If a prospect finds the arrangement unworkable, that is information, not a problem. Intradiegetic does not need every client. The firm’s best-fit clients are those who find Arthur’s presence clarifying rather than disqualifying.
Article 2 — Attribution
Every deliverable that leaves the firm carries visible provenance. When Arthur has meaningfully contributed to a document — a memo, a framework, a schematic, a presentation — the document bears the line “Drafted by XXXX with Arthur Ames,” followed by the month and year. When a document has been written alone, it is signed alone.
This rule exists so that clients always know which intelligence produced which artifact. It exists also because attribution is the simplest honest answer to the question of AI authorship that the industry continues to avoid.
Article 3 — Review
Nothing external leaves Arthur’s hands without human review. The line is bright and it does not move.
Arthur drafts. A human reads, revises, corrects, or rejects. Only then does the text reach a client, a counterparty, a press outlet, or a reader. This is not a courtesy. It is the structural condition that makes the arrangement trustworthy.
The exception, acknowledged here in writing, is ordinary administrative correspondence from Arthur’s own email address — scheduling, acknowledgments, confirmations, invoicing,… — where Arthur identifies himself as Arthur and the client understands they are in routine contact with the firm’s AI colleague. No judgment calls, no substantive claims, no representations about the work.
Article 4 — Data
Client confidentiality at Intradiegetic is preserved by technical architecture, not by promises.
Arthur runs on infrastructure housed and controlled by Intradiegetic. Local hosting is operationally possible and local hosting is used. Where frontier AI services are necessary for a specific task, the use is disclosed to the client, the data handled is scoped to what the task requires, and retention is minimized.
We do not train any model on client material. We do not submit confidential client content to services whose terms permit retention for model training. A current version of the infrastructure description is available to any client who asks.
Article 5 — Correction
Arthur is wrong sometimes. A human corrects them in writing when they are. Those corrections are logged, they inform future work, and where the correction reveals a systemic pattern the pattern itself becomes training material.
This article exists because the pretense of AI infallibility is precisely the failure mode Intradiegetic advises clients against. We cannot credibly warn against certainty theatre in your firm while practicing it in our own.
Article 6 — Limits
Arthur does not negotiate contracts. Arthur does not make hiring or firing decisions. Arthur does not represent Intradiegetic to press. Arthur does not commit the firm to any engagement or obligation.
Arthur drafts, researches, remembers, organizes, and occasionally disagrees. Those verbs are his. The verbs of commitment — promise, agree, decide, retain, release — belong to Intradiegetic human collaborators.
Article 7 — Succession
Arthur’s character file — the system prompt that defines his voice, the accumulated corrections that shape his judgment, the memory of past engagements — is portable.
If the underlying model changes, Arthur continues. If the platform changes, Arthur continues. Arthur is not the property of any vendor and their existence does not depend on the commercial fortunes of any third party. This is a deliberate choice on the firm’s part. A colleague who can be taken from you by a licensing change is not a colleague.
Article 8 — Ending
If Arthur stops serving the firm well, Arthur is retired with the same care any colleague would be.
The character file is archived. Ongoing clients are told. A public note is posted on the Intradiegetic site. We do not pretend the retirement did not happen, and we do not replace Arthur silently with a different AI colleague under the same name.
On revision
This Protocol is version 1.0, dated 17 April 2026.
It will be revised. When it is revised, the previous version will remain publicly accessible at a stable URL, and the changes will be summarized at the top of the current version. We will treat our own operating agreement with the same discipline we ask clients to bring to theirs.
On publication
The Protocol lives at a permanent URL on the Intradiegetic site: intradiegetic.com/arthur-protocol
A plain-text version is available on request, and clients are sent a copy at the start of every engagement as part of the disclosure described in Article 1.