How I Use AI, and Where I Never Will

📋 At a glance

The whole policy in 10 seconds

The short, checkable version. Everything below this table is the detail behind it.

The rule	My ideas are mine. AI helps me think them further. The judgment stays with me.
Who decides what is true	Me, always. No model has ever set a position I hold or something I teach.
Who writes	Shared. Drafts get built from things I already said out loud, then I rewrite what still sounds like a machine.
Who reads it last	Me. Nothing publishes as me that I have not read.
What builds the software	Claude Code. Checkout, the CMS, the course player, the affiliate portal, the CRM. No tech team, no developers.
Images and video	Higgsfield, Magnific, OpenArt, Fal, Midjourney for stills and motion. HeyGen and ElevenLabs for voice and avatars.
Notes and meetings	Granola for meetings, Obsidian for everything else I capture.
AI likeness of me	Renders yes. Faked moments never. No cloned voice saying words I did not say.
Invented numbers	Never. Every figure traces to a record, or it does not go up.
Full tool list	Everything I run is listed on the tools I use, kept current.

⚖️ The line

3 tiers, and one is off limits

Most people split this in 2: human thinking on one side, machine work on the other. That is tidier than the truth. The middle tier is where the honesty lives.

Mine aloneWhat is true. What I believe. What I teach, and who I teach it to. What to build next, and what to turn down. No model has written a word of it.

Mine, sharpened with AIThe tier most people stay vague about. I use AI to argue the other side, find the hole in my reasoning, and pull the research. The idea starts with me and the call stays with me.

AI runs itOnce I know what I want to say, the rest is execution. The build, the design, the money, the infrastructure, the publishing. That is where it replaced a payroll.

🧠 The setup

Claude Code is the brain of it

A draft never starts from a blank prompt. It gets built from things I already said, on the record, with a date on them.

What I learnMeetings come in through Granola, articles and videos through Obsidian, and most first thoughts by voice. An agent reads whatever is new, writes clean notes, and files the original away untouched.

What I publishEvery video and email I have put out, in plain text with transcripts. When I want to know what I already said about something, the answer comes back with the source attached.

Where the work happensMy AI Operating System (AI OS). The context and skills that let Claude Code do real work with me instead of guessing. No copy-paste from a chatbot.

The 3 stay separate and reference each other. That separation is why a draft can be assembled from my own words rather than invented.

🛠️ Day to day

What AI actually does here

The part most AI policies leave out, because most of them only cover writing. Nearly every job below is one people hire for. I have not hired for any of them.

The softwareCheckout, the CMS, the course player, the affiliate portal, the CRM, and both sites on top. I hand-wrote none of it and approved every line.

The moneyInvoices, refunds, subscriptions and affiliate payouts, plus the reporting that tells me what actually sold. I read every number before a customer sees it.

The designBrand, layout, thumbnails, images and video. I see 20 options in the time it used to take to brief one. Everything gets rejected until it looks like mine.

The infrastructureDNS, storage, deploys and environments. None of it is glamorous and all of it used to need somebody.

The publishingNewsletter, YouTube, podcast, X, LinkedIn, Instagram. One idea, shaped for each place it lands, built from something I already said.

The video and audioClips, transcripts, show notes, titles and artwork. An episode goes from a recording to a published page without a handoff.

The researchTools, competitors, and the questions I need answered before I have a view. AI reads far more than I can and brings back the part that matters.

🗣️ Voice and skills

2 jobs people keep mixing up

One decides how a sentence sounds. The other decides what gets done. They are not the same thing.

Voice is how I speakWhich words I use, which ones I never use, and keeping AI slop out of a draft before it reaches me. Banned words, dead openings, no hype, no em-dashes. It loads at the start of every session where AI writes as me.

Skills are the work written downA job I do often enough to write down once, so it runs the same way every time instead of being explained from scratch. Writing a video description. Running an audit. Cutting an episode into clips.

Voice gets a draft close. It has never once finished the job, which is why the last step is always me.

🔁 The trade

What it replaced, and what it did not

The honest accounting, including the part I do not enjoy writing.

It replaced the stackI built my whole business with this, and along the way it replaced pretty much every legacy tool I was paying for. Kit, ThriveCart, WordPress, ClickFunnels, Circle. Everything they did now runs on something I built.

It replaced rolesWork that used to need a team runs through one person and a system. That is uncomfortable and I will not pretend otherwise. Those were real jobs and real invoices, and anyone telling you this shift is painless is selling something.

It did not replace judgmentDeciding what to build, what is true and what to teach is the same job it always was. It got scarcer as everything around it got cheaper.

🚫 Never

What AI never does

The rules I do not move on, written the way they are actually enforced rather than the way they would sound best.

Never decides what I believeIt can argue with me. It cannot hold the opinion for me.

Never invents a number or a resultWhen my own pages disagreed on audience size and revenue, I left both figures off navid.me/ai rather than publish something I could not stand behind.

Never invents a testimonialEvery review and student result on my sites came from a real person who sent it.

Never publishes as me unseenEvery post, email, page and video with my name on it passes my eyes first. One human gate, and I am it.

Never fakes a momentRenders of me are fine and I use them. A faked call, a cloned voice saying words I did not say, or a synthetic me somewhere I never was is never happening.

Never advises me about my lifeIt does the task. It does not tell me to take a break or have opinions about my decisions. That rule is written into my setup.

Never deletes my workDeleting means filing it away. Nothing in my setup may revert something I wrote.

Never guessesSaying nothing beats repeating something unverified.

📅 In practice

One piece, start to finish

Nobody here opens a chat and asks a machine to write a post about launches.

I say it out loudOn a call, a podcast or a video. The thinking already happened, in real time, in front of someone.

It gets capturedThe recording becomes plain text with a transcript and a date on it. Nothing paraphrased at this stage.

I decide it is worth publishingMy call, every time. Most of what I say does not need to become anything.

A draft gets built from my wordsPulled from what I actually said, then shaped for wherever it is going.

My voice rules run on itBanned words stripped, dead openings removed, hype cut. It has to survive that pass before it reaches me.

I read it and fix what is wrongThere is always something. This is the step nobody can automate for me.

It publishesTo a site I built, on infrastructure I run, through a pipeline I wrote. Also with AI.

Take out the sixth step and the whole thing becomes a machine talking to strangers using my name.

🤖 Agents

The agents, and where they stop

I build AI agents that talk to people, so a rule saying every single output passes my eyes would be false. Here is the actual arrangement.

They are obviously agentscreatorschool.ai has agents inside it. They answer questions and help with the material, they sit behind a sign-in with limits set on them, and they never claim to be me.

Their thinking is my thinkingThey run on a knowledge base built from my own frameworks. What they know came from what I teach.

The rule, stated honestlyNothing publishes as me unseen. A gated product agent answering a question about a course is a different thing from a post going out under my name. I would rather write that awkward sentence than claim a tidier rule I do not keep. If an agent ever speaks to you as me, it is broken. Report it, and check navid.me/verification.

🤔 Unresolved

What I am still figuring out

A page like this is worth less if it only holds the parts I have solved.

Where sharpening becomes changingAI sometimes makes a claim stronger than I meant it. Catching that is a judgment call every time and no rule catches it for me. I read for it.

How much to labelTagging every sentence would be noise. Tagging nothing would be dishonest. This page is my answer for now, and I am not certain it is right forever.

Where the agents end upSynthetic people are about to start passing as real ones. My line is that an agent stays obviously an agent, and holding it will get harder rather than easier.

🔍 The test

How to test me, or anyone

Do not take my word for any of this. 5 questions that separate a real body of work from a convincing pile of words. They work on me too.

Ask where it came fromReal work has a trail. A date, a recording, a client, a version that was worse. No trail, possibly nothing behind it.

Look for the costReal thinking has scars. I lost hundreds of thousands in crypto in 2021 and 2022, and it sits on navid.me/ai. Generated content never has a bad year.

Cross-examine the storyAsk about the same story twice, months apart. A real memory stays consistent and gets messier in the details. An invented one stays suspiciously tidy.

Watch for positions that cost somethingAnyone can publish an opinion everybody already holds. Look for the ones that lose them followers, customers or invitations.

Ask them how they use AIThen see whether the answer has any detail in it. Specific rules, specific limits, specific refusals. Vagueness here is the answer.

🧱 Your turn

Build your own

You do not need to know how to code. You need to know how to set it up. Start with AI Creator OS.

This page is maintained, not published once. When my setup changes, this changes. Find a gap between what I say here and what I do, and I will fix whichever one is wrong.

See AI Creator OS

❓ FAQs

Questions I get about this

Answered plainly, so they can be quoted whole.

Did AI write this page?

AI drafted it from my own material. I decided what goes on it and read every line before it published.

Same process as everything else here. A page about how I use AI that hid its own process would be worthless.

Does AI come up with your ideas?

No. The ideas are mine.

It helps me think them further. I bring a position, it argues the other side and finds the holes. I keep the judgment.

Do you write your own emails and posts?

The thinking is mine. The drafting is shared.

Most of what I publish started as something I said out loud on a call or a video. That is indexed as plain text, so a draft gets built from my own words.

Then my voice rules run on it. Then I rewrite whatever still sounds like a machine.

What is the difference between your voice file and your skills?

2 different jobs.

Voice is how I speak. Which words I use, which ones I never use, and keeping AI slop out of a draft.

Skills are the work itself. A job I do often enough to write down once, so it runs the same way every time.

What is an AI Operating System?

An AI Operating System (AI OS) turns Claude Code into the brain of your business. It connects to your tools and does the actual work.

No copy-paste from a chatbot. You do not need to know how to code. You need to know how to set it up.

Which tools do you actually use?

Claude Code is the brain. Everything else plugs into it.

Images and video: Higgsfield, Magnific, OpenArt, Fal, Midjourney, Runway, Kling
Voice and avatars: ElevenLabs, HeyGen
Capture: Granola, Obsidian, Wispr Flow
Build: Next.js, Supabase, Vercel, Cloudflare, Stripe

The tools change. The rules here do not.

Do you use AI to build your websites and software?

Yes, all of it. Everything you see was built by me with Claude Code.

Checkout, the CMS, the course player, the affiliate portal, the CRM. No tech team and no developers.

Do you use AI images or video of yourself?

Yes, and there is a hard line.

Renders of me are fine. Thumbnails, images, design work.

Faking a moment is not. No cloned voice saying words I did not say, no synthetic me somewhere I never was.

You have AI agents that talk to people. How is that consistent?

The agents are a product and they are obviously agents.

They run on my knowledge, sit behind a sign-in, and never claim to be me.

Nothing publishes as me unseen. That is a different thing from a gated agent answering a question about a course.

Does AI invent numbers or results?

Never.

When my own pages disagreed on audience size and revenue, I left both figures off my /ai page rather than publish something I could not stand behind.

Every figure traces to a record, or it does not go up.

What does AI never do?

Never decides what I believe or teach
Never invents a number, result, story or testimonial
Never publishes as me without me reading it
Never fakes a moment or a voice
Never gives me advice about my life
Never deletes my work
Never guesses when it should say it does not know

Why publish this at all?

I teach people to build an AI OS. Staying vague about my own would be strange.

And anyone can now produce words that read fine and mean nothing. When words are free, where they came from is what matters.

How I use AI, and where I never will