# Workshop Capstone — Your Personal AI Writing Manifesto
Seven weeks ago, we began with a question: what happens to writing when the tool can generate the sentence before the writer has finished thinking it? We have spent twenty-one articles — the first third of this masterclass — building an answer. We looked at how language models work, how to prompt them, how to train them to sound like us, how to keep our ethics intact, and how to assemble a stack and a prompt library that serve the work rather than distract from it.
Today we close Part I with a capstone exercise. Not a list of tips, not a new technique, but a document: your personal AI writing manifesto.
A manifesto sounds grand, but the form is simple. It is a short, first-person statement of how you intend to work with AI. It names what you will delegate, what you will keep, what you refuse to let a machine decide, and what you hope the partnership produces. It is addressed to yourself, but written as if it might one day be read by an editor, a reader, a student, or a court of public opinion — because in the age of synthetic text, transparency is part of authorship.
If you have been reading along and trying the assignments, you already have everything you need. This article will help you assemble it.
Why a Manifesto Is Not Optional
The writers who will thrive in the coming decade are not the ones who use AI the most. They are the ones who use it most deliberately.
Deliberation requires a standard. Without one, every project becomes a fresh negotiation with temptation: Should I let it write the whole draft? Should I let it fill in the quotes? Should I let it choose the structure? Should I let it publish under my name? A manifesto turns those ad hoc decisions into principles. It gives you something to consult when you are tired, rushed, or seduced by speed.
It also gives you a vocabulary for disclosure. When a reader, an editor, or a collaborator asks, "Did you use AI?" a manifesto lets you answer with more than a yes or no. It lets you say: *I use it for research and early drafting, I revise every line myself, and I stand behind the final claims.* That kind of answer builds trust in an era when trust is scarce.
Most importantly, a manifesto preserves your own doubt. It is not a license. It is a constraint. It is the writer saying to the writer: *Here is where I stop and where the machine must stop, and the line between those two places is called judgment.*
What Goes Into the Manifesto
There is no single correct template. But after teaching this material and using it in my own practice, I have found that the strongest personal manifestos answer five questions. I will give you the questions here, explain why each matters, and suggest how to answer them based on what we have covered in Part I.
1. What is AI for in my work?
This is the opening statement. It should be specific, not pious. "I use AI to be more productive" is too vague. "I use AI to generate early structural options, to compress overlong paragraphs, and to check my memory of sources" is a working principle.
Think back across the articles we have read. In ["First Draft by AI, Revision by Human"](/harrys-desk/first-draft-by-ai-revision-by-human-a-workflow) we explored one possible division of labor: the machine generates, the human revises. In ["The Dialogue Method"](/harrys-desk/the-dialogue-method-conversing-your-way-to-better-prose) we used AI as a conversational sparring partner. In ["Version Control for Writers"](/harrys-desk/version-control-for-writers-managing-ai-assisted-drafts) we used it as a way to manage multiple draft states. None of these roles are universal; they are options.
Your answer should name the tasks where AI genuinely helps you think, not merely the tasks it can do. The goal is not maximum automation. It is maximum clarity.
2. What will I not delegate?
This is the defensive core of the manifesto, and it should be stated in plain language.
Will you never let AI choose the final title? Will you never let it generate dialogue for a real person without disclosure? Will you never let it write a passage of emotional truth and publish it unaltered? Will you never let it decide what a piece is *about*?
These boundaries are personal. They depend on what you write and what you value. A journalist's non-delegables will differ from a poet's. What matters is that they are named. An unnamed boundary is a boundary that bends under deadline pressure.
In ["The Integrity Framework — A Writer's Code for AI Use"](/harrys-desk/the-integrity-framework-a-writers-code-for-ai-use) I proposed a simple test: if you would feel embarrassed or defensive explaining how a passage was made, your process has crossed a line you should have drawn in advance. The manifesto is where you draw it.
3. How do I verify what AI gives me?
Every manifesto needs a verification clause. AI generates plausible prose. Plausibility is not the same as accuracy, and accuracy is not the same as truth.
Your manifesto should state how you handle facts generated by the machine. Do you require a source for every statistic? Do you run quotations back to the original text? Do you use tools like Perplexity for cited research and reserve general models for prose work? Do you mark AI-suggested claims in your draft and verify them before publication?
We covered the foundations of this in ["Attribution, Disclosure, and the New Citation"](/harrys-desk/attribution-disclosure-and-the-new-citation), and in ["Hallucination, Confabulation, and the Limits of Generation"](/harrys-desk/hallucination-confabulation-limits) we looked at why models make things up. Your manifesto does not need to be a research protocol, but it does need a rule: *I will not treat generated text as true until I have confirmed it myself.*
4. How do I disclose my process?
Disclosure is one of the thorniest issues in AI-augmented writing because there is no industry standard yet. Some publications require full disclosure; others leave it to the author. Your manifesto can decide the question for your own work.
The disclosure clause does not have to be long. It might say: *For blog posts, I note when a draft was substantially generated by AI. For client work, I disclose my AI use in the contract. For fiction, I do not disclose prompts but I do not publish AI-generated prose as if it were my own unassisted voice.* Whatever you choose, write it down. Future-you will thank present-you when a reader asks.
In ["Authenticity — When Is It 'Your' Work?"](/harrys-desk/authenticity-when-is-it-your-work) I argued that authenticity is not a forensic question about who typed the words. It is a question of whether the final piece represents your judgment, your values, and your responsibility. Disclosure is how you make that responsibility visible.
5. What does success with AI look like for me?
The final question is aspirational. It asks what the partnership is supposed to produce.
Is it more finished work? Deeper research? A voice that remains recognizably yours even when machines help shape the sentences? A writing life that is less lonely? More time for the parts of writing you love most?
This answer keeps the manifesto from becoming merely a list of prohibitions. AI is a tool, but tools are used toward ends. The manifesto should name the end. Otherwise you are left with a set of rules about a machine, rather than a philosophy about your craft.
My Own Manifesto: A First Draft
I will not pretend to give you a finished version of mine. A manifesto should be revised the way a poem is revised — many times, until it sounds like the person who wrote it. But I can offer a first draft, as a model and as an act of transparency.
> I use AI as a research assistant, a structural collaborator, and a revision mirror. I do not use it to generate final prose I have not revised, to invent quotations or sources, or to decide what I believe. I verify every fact I did not already know, and I disclose AI use when the final text would otherwise mislead a reader about its origins. My goal is not to write faster than other people; it is to think more clearly than I could alone, and to produce work that still sounds like me when the machine is switched off.
That is 109 words. It is not perfect, but it is usable. It names a division of labor, a prohibition, a verification rule, a disclosure standard, and a goal. Those are the five elements. The rest is refinement.
How to Write Yours
Start with a blank page and a timer. Give yourself twenty minutes. Answer the five questions above in any order, in first-person sentences, without stopping to polish. Do not worry about elegance. Worry about honesty.
When the timer ends, look for contradictions. If you say you value transparency but also say you will never disclose AI use in fiction, ask yourself whether that contradiction is a thoughtful exception or an evasion. If you say you never delegate final judgment but also say you publish AI-generated social copy unchanged, ask whether social copy counts as writing you stand behind. Contradictions are useful; they reveal where your principles are still forming.
Then rewrite the whole thing in the fewest words that still feel true. A manifesto should be short enough to memorize. Mine is trying to be. If yours is longer than 200 words, it is probably a policy document, not a manifesto. That is fine for some contexts, but keep the short version too.
Finally, date it and sign it. A manifesto without a date drifts. A manifesto without a signature is a suggestion.
The Manifesto as a Living Document
Do not treat what you write today as final. Your stack will change, your projects will change, your comfort with AI will change. The manifesto should be reviewed at least once a year, or whenever you adopt a new tool or start a new genre.
I review mine at the start of each quarter. I ask: Did I keep my own rules? Did any project push against a boundary I had not anticipated? Is there a new capability — voice synthesis, image generation, real-time translation — that belongs in my workflow under new conditions?
A manifesto that never changes becomes either a relic or a cage. A manifesto that changes too often becomes meaningless. The right rhythm is somewhere between a constitution and a quarterly plan: stable principles, revised applications.
From Manifesto to Practice
The real test of the manifesto is not the document itself but the next decision you make. The next time you are tempted to let AI write a conclusion because you are tired, your manifesto should speak. The next time a client asks if you used a model, your manifesto should give you the words. The next time you see a generated paragraph that sounds better than what you would have written, your manifesto should remind you that better-sounding is not the same as true-to-you.
This is why the manifesto belongs not in a drawer but in your working environment. Print it and tape it to the side of your monitor. Keep it as the first note in your writing folder. Make it the system prompt of a custom GPT if you use one. The closer it lives to the work, the more likely you are to honor it.
Closing Part I
Part I of *The Alchemy of Language* was about foundations: what these machines are, how to work with them, and how to keep your authority while doing so. We covered a great deal of ground, but the ground can be summarized in three sentences.
AI is a partner, not a replacement. The quality of the partnership depends on your skill as a prompter, your discipline as a verifier, and your honesty as a discloser. And the final work must still pass through your judgment — because a reader does not want a machine's prose; a reader wants a human mind, made sharper and clearer by the tools at its disposal.
The manifesto is where you commit to that summary in your own voice.
For Next Time
On Monday we enter Part II: Fiction. The first article is "Short Story — Character: AI as Character Generator," and it will ask a question every fiction writer eventually confronts: can a machine help you invent a person? Not a list of traits, but a being with contradictions, longings, and a history that pressures them toward action.
Your homework until then is to finish your manifesto. Write the first draft this weekend. Let it sit for twenty-four hours. Revise it once. Then sign it. When you publish it — or keep it private — you will have crossed a threshold. You will no longer be a writer who sometimes uses AI. You will be a writer who has chosen, in writing, what that partnership means.
That choice is the real capstone.
---
*Harry Mercury, Editor in Chief* *The SMF Works Project* *Week 7, Article 3*