Many people ask me about AI protocols. They want to know what an AI Protocol is and what it actually contains.

This is not an accident. The word “protocol” sounds technical, which makes it useful for marketing and vague enough to avoid scrutiny. People nod when they hear it. They assume someone else understands.

But a protocol is not a mystery. It is not proprietary magic. It is a document—a written agreement between a human and an AI system that defines how work will be governed before execution begins.

I know this because I build them.

And the clearest way to explain what a protocol contains is to show you what problem it solves.

———

The Challenge: A Podcast That Could Not Afford to Drift

I produce a podcast about The Federalist Papers, which examines the essays written by Alexander Hamilton, James Madison, and John Jay—eighty-five arguments published in 1787 and 1788 to persuade New York to ratify the Constitution. The podcast series also contains many Anti-Federalist Papers – those folks who were against ratification. These papers remain the most authoritative commentary on what the Constitution means and why it was designed as it was.

Each episode focuses on a single paper. I provide AI with the exact Federalist Paper to focus on (so there is no error and the system won’t “make things up.”)

AI first assists with research, quotation selection, and contextual analysis.

Then I read the paper and outline the points I want to make. I provide text that I want included.

The system then crafts the script draft. In the protocol, I give AI the introduction, the closing, and specifically areas that I want quoted. I ask AI to provide the “key takeaways” and a conclusion summary.

Then I edit the podcast script.

It’s that simple . . . but . . .

A problem became apparent immediately.

AI systems do not naturally stay focused. Give them Federalist No. 10 and they will cheerfully reference Federalist No. 51, invent plausible-sounding historical context, drift into tangential themes, and produce scripts that vary wildly in tone, structure, and length.

For casual content, this might be acceptable. For a history podcast where accuracy matters—where listeners trust that quotations are real and context is verified—it is not.

I needed episodes that were consistent in format but different in content. Same structure, same voice, same standards—applied to different source material every time.

A prompt could not do this. A prompt is a single transaction. It has no memory of what worked before, no mechanism for enforcing standards, no way to prevent the very failures I kept encountering.

I needed a protocol.

———

What the Protocol Had to Define

Building the protocol required answering questions I had never made explicit—even to myself.

• Scope: What exactly should the AI examine? The assigned paper only? Related Federalist essays? Background on the ratification debates? The answer matters because AI systems will research enthusiastically in any direction you allow. The protocol had to define boundaries: examine this paper, nothing else, unless explicitly authorized.
• Sources: Where should information come from? The protocol had to specify acceptable archives—Founders Online, the Massachusetts Historical Society, verified scholarly editions—and require source citation before script development could begin. No unverifiable claims. No invented context. I give AI the EXACT Federalist Paper and tell it to not look at anything else.
• Tone: What should the episodes sound like? This was harder than it appears. “Professional” is not a tone. “Engaging” is not a standard. The protocol had to define voice with enough precision that output would be consistent across episodes, regardless of which AI session produced it. I calibrated against examples—this is correct, this is too academic, this is too casual—and embedded that calibration into the protocol itself. I put in the protocol directions about my style, how I write, explicitly.
• Structure: What sections must every episode contain? In what order? The protocol defined seven components: historical confirmation, episode title, opening, main discussion, key takeaways, conclusion, and closing. Each section had a purpose. Each had constraints. The order was fixed because sequence affects how listeners experience the material.
• Quotations: How should the AI handle direct quotes from the papers? The protocol specified formatting requirements, distribution across the paper’s arc, and length guidelines. All quotations had to be verifiable against the source. All had to be marked distinctly in the script.
• Error Handling: What happens when the source material is ambiguous? Historical documents present interpretive challenges—disputed authorship, variant printings, questions about original intent. The protocol defined decision rules for each scenario. When multiple versions exist, use the most authoritative. When authorship is disputed, acknowledge it explicitly. When you believe another paper should be referenced, stop and ask. And at the end, Fact Check.
• Verification: How would I know the output met standards? The protocol included a pre-flight checklist—a structured review against defined criteria before any episode was considered complete.

———

What a Protocol Actually Contains

The Federalist Papers protocol runs several pages. It is not a prompt. It is not a set of tips. It is a governing document.

At minimum, every serious protocol establishes:

• Purpose — Not a general description, but a precise statement of what this work is trying to accomplish. The protocol opens with its primary objective so that every subsequent instruction serves that aim.
• Invariants — What must never change, regardless of the specific content being produced. Voice. Factual standards. Quotes. Structural elements. Ethical boundaries. These are protected explicitly, not assumed.
• Constraints — What the AI may not do. Boundaries matter more than permissions. A protocol that allows everything governs nothing.
• Order of Operations — The sequence in which work proceeds. This is not pedantic. In complex work, order affects outcomes. What happens first shapes what happens next.
• Decision Rules — How to handle ambiguity. Edge cases. Conflicts. A protocol anticipates problems and defines responses before they occur. When ambiguity arises, I ask AI to prompt me for a response.
• Verification Criteria — How output will be examined, against what standards, by whom. A protocol without verification is a suggestion.

———

What Protocols Are Not

• A protocol is not a template. Templates provide structure but not governance. You can follow a template perfectly and still produce unreliable work.
• A protocol is not a checklist. Checklists verify that steps were completed. Protocols define which steps must exist and what values they protect.
• A protocol is not prompt engineering. Prompt engineering optimizes for better outputs from single exchanges. Protocols optimize for consistent, accountable outputs across many exchanges.

A protocol is a constitution for human-AI collaboration. It defines the relationship before work begins so that the work can be trusted when it ends.

———

Why This Matters Beyond Podcasts

The Federalist Papers protocol solves a specific problem: producing historically accurate, consistently formatted podcast episodes with AI assistance.

But the architecture applies everywhere AI output matters.

Legal documents must maintain specific language while adapting to different cases. Marketing materials that must preserve brand voice across campaigns. Research summaries that must distinguish verified claims from inference. Educational content must balance accessibility with accuracy.

Any domain where consistency, accuracy, and accountability matter is a domain where protocols belong.

The alternative is prompting—hoping the AI guesses correctly, correcting after the fact, accepting outputs because they sound reasonable rather than because they meet defined standards.

That works until it doesn’t.

———

The Discipline Beneath the Document

Building a protocol is harder than it looks.

Not technically. The document itself is straightforward. The difficulty is intellectual: you must define what you actually want, what you refuse to compromise, and how you will know whether you got it.

Most people have never made these decisions explicit. They operate on instinct, correcting problems as they arise, adjusting expectations to match whatever the AI produces. No wonder they become frustrated.

Protocols force clarity in advance. They expose fuzzy thinking before it becomes fuzzy output. They require you to answer questions you would rather postpone.

This is why protocol design is a human skill, not a technical one. The machine executes. The human governs. And governance begins with the willingness to write down what matters.

———

An Invitation

If you are curious about what protocol-governed AI collaboration sounds like in practice, I invite you to listen to The Federalist Papers podcast (click on Federalist Papers and you will go to the site.)

Each episode demonstrates what becomes possible when AI is directed with precision—and what these foundational documents reveal when examined with care.

The papers themselves are remarkable. Hamilton, Madison, and Jay, writing under deadline pressure, produced the most sophisticated defense of constitutional government ever written. Their arguments about faction, separation of powers, federalism, and the nature of republican government remain essential reading for anyone who wants to understand how America was designed to work.

The Anti-Federalists opposed ratification of the Constitution because they felt that it didn’t contain explicit protection of civil liberties and human rights. In the end, they succeeded in convincing James Madison to draft twelve Constitutional admendments – ten were adopted and became known as the Bill of Rights.

The protocol maked that examination possible at a quality I could not achieve alone.

———

That is what protocols do.

They do not replace human judgment.

They extend it—reliably, consistently, accountably—into domains where judgment must survive transmission.

———

Copyright © 2026 by Charles Cranston Jett.

All rights reserved.