How I Use AI

The charge used to be “you’re wrong.” Now it’s “you’re not real.”

Someone accused me again this week of using AI to write my content. They meant it as an insult. The accusation runs like this: the writing is too dense, too fluent, too consistent, too fast. Nobody produces this much material at this level without a machine writing it. Therefore, machine. One accuser this round conceded the point was good in the same sentence he dismissed the writing as machine-made — he took the argument and denied the author in one line.

What that reads as, once you run it through, is this: the accuser doesn’t have a model for what forty-five years of accumulated framework work looks like when the operator finally opens the taps. I’ve been in the restaurant business since April 4, 1982. Twenty-two of those years running operations directly. The rest consulting with operators. Somewhere in the neighborhood of 150 restaurant openings. Every one of those years I was taking notes, writing articles, drafting frameworks, naming patterns, and building the vocabulary I now use to teach the operating physics of this business. The corpus was already there. What changed is when I decided to publish at the pace the framework had been accumulating at for four decades.

That’s not AI writing. That’s an operator running a refinement pass on his own body of work.

The Record Is The Proof #

There is a simpler answer than any argument about process, and it is the one the accuser can check. Every position I take traces to something I already published, with a date on it. Every piece I write closes with a Digging Deeper section that links the positions it stands on. Every term I use is defined in my Knowledge Base, and every entry there carries its date.

A claim about how I work asks you to trust me. The record does not ask anything. It lets you check. The positions were on the record before the tools in question existed, and the forty-five years they came from were spent in restaurants, not in a chat window.

That matters more this year than it did last year. The major AI companies now mark the text their models produce: an invisible statistical pattern in word choice, put there to meet the transparency rules in the European Union’s AI Act. Anthropic marks the text Claude writes. xAI begins marking the text Grok writes on November 7. Text a model drafts carries that pattern wherever it is pasted.

A watermark can tell you who chose the words. It cannot tell you whose idea the words carry. Only the record can do that, and mine is public.

The Charge Used To Be “You’re Wrong” #

The old charge engaged the argument. “You’re wrong” was falsifiable, debatable, answerable. You could meet it with evidence, with positions on the record, with decades of receipts. That’s a fight an operator takes.

The new charge dismisses the speaker. It doesn’t engage what you wrote — it denies that you wrote it. You cannot produce a birth certificate for a sentence. You can produce the record.

Understand what actually changed: the accusation didn’t get smarter. It gave up. “You’re wrong” at least assumed you existed. “You’re not real” is what the charge becomes when the accuser has run out of ways to be right.

Human Slop #

Slop is not an AI phenomenon. Humans have been producing it since the dawn of time — the ghost-written executive byline, the committee-written mission statement, the formulaic press release, the keynote nobody wrote and nobody read. Long before any machine generated a sentence, mediocrity was fully industrialized.

When a reader calls writing “AI slop,” they are describing slop and attributing it to the newest tool in the room. The tool didn’t bring the slop. The slop was already ours.

And when everything reads like slop to you, that is not a fact about the writing. That is a fact about your ear. The machine already won your reading.

There Is No Unmediated Expression #

Here is the part the accusation cannot survive: there is no act of expression that isn’t tool-mediated. The voice is a biological tool — vocal cords, breath, mouth. The hand holding the pen is a tool. Expression is the act of putting thought into a medium, and a medium is an instrument. The only unmediated thing is thought that never leaves your head, and that isn’t expression. That’s privacy.

So the accuser’s line was never “tool versus no tool.” It’s “old tools versus new tools” — voice and hand feel authentic because they’re the tools this era is comfortable with, the same way a pen once felt authentic next to a typewriter. The distinction collapses to comfort. It was never about authenticity, because the authenticity claim was never available to make.

Run the logic one step further and it demands a return to cursive — the hand, the older tool, therefore the more authentic one. One step past that, the quill. One step past that, the cave wall. The argument cannot say where authenticity stops, so it stops wherever the accuser happens to stand. That is not a position. That is a footprint.

I read the accusation on my pocket computer and answered it on another computer. Some of the math I do on a calculator. The accusation itself reached me because a feed algorithm decided I should see it, on a platform that until last month offered to write the post for you and now offers to proofread it.

Now run the rest of your day. The alarm that woke you was a computer. So was the coffee maker that started on a timer, the thermostat that controls the temperature in your house, the meters that read how much electricity and water you use at home and at the office, the router that counts every byte of wifi you pull, the doorbell camera that watched you leave, the car with more computing in it than the machines that sent men to the moon, the navigation that rerouted you around the wreck, the toll tag that paid on the way, and the radio that picked the next song. At work, the badge opened the door, the inbox filtered the spam, the calendar found the meeting time, the spell-check fixed the memo, the spreadsheet ran the formulas, and the video call blurred the room behind you. At lunch, the app found the restaurant, the reservation platform held the table, the host stand quoted the wait, the tablet fired the ticket, the kitchen display timed the fire, the combi oven held the temperature, the point-of-sale system split the check, the card reader ran the payment, and the software calculated the tip. On the way home, the grocery app built the list, the self-checkout scanned the cart, and the bank app deposited the check from a photograph. That night, the streaming service chose the show, the smart speaker set tomorrow’s alarm, the refrigerator reported the filter, the watch counted your steps and scored your sleep, and the phone finished every sentence you typed. When something went wrong with any of it, you talked to a bot before the brand would let you talk to a person.

Every tool in every day of our lives does part of the work. Nobody calls an operator’s food cost fake because a calculator ran the division. Nobody says you didn’t do the dishes because the dishwasher ran the cycle. Nobody calls your commute fake because a computer picked the route. The question was never whether a machine touched it. The question is who decided. You are reading the answer on a computer.

Why The Pace Changed #

I did not always write at this pace. Before COVID, I documented my work in fits and starts. I published when I had time. I let entire framework decisions sit as notes in a folder because I assumed I had time to formalize them later.

Then I lost my son, my mother, and my father in the same year.

That year rearranged my read on time. I decided that whatever life I had left, I was not going to spend it on superficial arguments that never went anywhere. I was not going to spend it in street fights with people who were a waste of my time. I was not going to spend it hedging positions I already knew were correct to make industry counsel comfortable.

I decided my life’s work was worth enough to me that I owed it my time — and I owed my time to the operators my framework was built to help.

That is the decision that changed the pace. Not AI. A decision, made in the aftermath of loss, that my framework was going to get published and the arguments were going to get taken seriously by an operator who no longer had time for the alternative.

Why The Register Changed #

The other read on my writing right now is that it does not sound like most social media writing. That is correct. It doesn’t.

I’ve always been a serious, formal writer. That is my native register. But I learned at an early age that people don’t like smarter people, and I spent decades paying that tax. I dumbed down my thinking and my writing so the room stayed comfortable. The platform decade pulled the register down further — I got into fights I should not have been in, I engaged with commenters whose only goal was to be seen engaging with me, and I allowed the platform to shape my writing toward its mean, which is short, reactive, and driven by whoever was loudest that day.

Two things stopped the adjusting. A man I admire told me to stop it — that writing down to the room was the actual offense. And then I lost my son, my mother, and my father in the same year, and the question of whether to spend whatever life I had left adjusting for anyone’s comfort answered itself.

When I decided to publish my framework in full this year, I decided to publish it in the register the framework deserves — direct, load-bearing, unhedged, and grounded in decades of successful experience running and consulting to real restaurants.

I am educated. I have four decades of results. I have positions on the record that I built the hard way. I owe those positions the register they were built in. I do not owe the platform a softer version.

That is not AI writing. That is a decision about what my writing is for. And the “not real” charge is the price of un-adjusting. It arrives the moment the level comes back up.

What AI Actually Does For Me #

Here is what AI does in my workflow, precisely:

It reads the corpus faster than I can. When I’m drafting a new piece and I want to know whether I’ve already published a position on a specific point, AI can run semantic search across the documents in my Knowledge Base in seconds. That’s not writing. That’s retrieval.

It flags casing violations and vocabulary drift. My framework runs on a locked lexicon — specific bracketed IP terms with canonical casing that has to match across every surface. AI catches when I’ve written a term wrong. That’s proofreading, not authorship.

It surfaces family relationships between terms. When I’m working on a new dictionary entry, AI can pull every related, opposing, and parent term from the corpus and hand me a list of what needs cross-referencing. That’s indexing, not thinking.

It runs pushback against my drafts. When I want to know whether a position I’m taking will hold up under a specific counter-argument, AI can run the counter-argument at me hard. The counter-arguments are positions someone else has already taken (a study, a court ruling, an industry view) or a check on a fact or a number. I decide whether the position holds or needs revision. It argues. It does not vote. That’s editing, not writing.

It drafts from my workshops. I work a subject through, argue it, test it against the corpus and rule on every open question, and AI records what I decided in draft form. I correct the draft, cut it and approve it, or I throw it out. The decisions in the draft were made before the draft existed. That’s a recording secretary with an editor’s pen, not an author. A kitchen manager who writes the spec, sets the recipe and tastes every plate at the pass made that plate, even when a line cook fired it.

It converts formats. Markdown to HTML. Long-form to teaser. Article to KB entry to LinkedIn post. That’s typesetting, not creation.

Nobody carries a thesaurus in their head, or a dictionary. Writers have kept both on the desk for as long as there have been books, and nobody ever said the book belonged to Roget or Webster. Every book on every shelf also went through an editor who changed the author’s words, and some of those changes said the point better than the author did. Nobody takes the author’s name off the cover. The dictionary checks the word, the thesaurus offers another, the editor finds the better one, and the author decides. The point stays the author’s. Nobody announces the thesaurus or the dictionary. Nobody should have to.

What AI Cannot Do In My Workflow #

Every one of the above is a labor multiplier on work I’m already doing. None of them produces a framework. None of them writes a load-bearing piece. None of them has an operator’s read on a specific problem.

Here’s what AI cannot do:

It cannot name a mechanism that isn’t in its training data. Every load-bearing IP term in my framework — Operator Arbitrage, The Guest Contract, Road Cancer, Framework Arbitrage, Symbolic Price Equity, Edison Trust Arbitrage, Case Study Reduction — is a term I named because I saw the mechanism operating in real restaurants and no existing vocabulary named it correctly. AI can restate a term I’ve defined. It cannot originate one.

It cannot hold a position under contradiction. My framework takes positions that contradict industry counsel across half a dozen dimensions. Every one of those positions is a decision I made after running the mechanism against real operations for years. AI does not have those years. It has statistical adjacency to industry counsel, which is exactly the wrong data set for taking positions against it.

It cannot read a specific operation. My core consulting work is running framework-level reads on live operations. AI has never worked a shift. It has never watched a lead break under a Guest Contract violation. It has never seen a cast member fork. It has never sat across a table from an operator whose numbers are collapsing and whose Perspective is the actual constraint. That work is not writeable by anything that hasn’t done it.

It cannot generate coherence across the whole corpus. Every entry in my Knowledge Base cross-references every other relevant entry in the corpus. Every position I take on the blog reinforces positions I take in the books. Every diagnostic test in a fieldbook maps to a mechanism in the parent book. That coherence is not an AI output. That coherence is my framework itself, and it was built one decision at a time over decades.

The Card Box #

I did not build the record to answer accusers. I built it because I hold myself to my own work, and that means more than my positions. It means the dates, the facts, the figures, the quotes, the sources and everything else that sits next to the work. If I said it last year, I should still mean it this year, or I should say on the record why I changed my mind. If I cited a number, it should still check out. If I quoted someone, the words should be theirs, exactly as they said them. The only way to keep that promise across this many words is to catalogue all of it and check every new piece against what is already there.

That makes the catalogue a credibility tool, and it is one we did not have in the toolbox before. A writer could keep a card box, but no writer could check every new sentence against forty-five years of work before it went out. I can. The tool the accuser points to as the reason not to trust the work is the same tool that lets me prove it holds together.

Underneath the accusation sits a fair question: how does anyone who has written this much, on this many subjects, keep it consistent? The answer is older than any of these tools.

Niklas Luhmann, the German sociologist, kept a card index of roughly 90,000 handwritten notes, numbered and cross-linked, with a keyword index to find his way back through them. He credited it with making possible his roughly fifty books and 550 articles. Before he wrote something new, he checked it against what the box already held.

My Knowledge Base Dictionary is my style sheet: every locked term, checked before it is used. My corpus is my card box. The difference is the instrument. Luhmann wrote his cards by hand. I recorded mine on a computer and a calculator, and I use another computer to sort them, search them and edit what comes out of them. Nobody suggests his work stopped being his because a box held the notes.

His box could not argue with him. Mine can. It still does not get a vote. Where the corpus has nothing to say on a new question, the gap comes to me, and nothing goes on the record until I rule on it. The patterns in the record are mine. I recognized them over forty-five years of working in restaurants, and I wrote them down. AI finds where I wrote them. It does not put anything there.

The Real Read On The Accusation #

When someone accuses an operator or a consultant of using AI to write their content, one of three things is running underneath:

They don’t have a model for volume plus coherence. They see the pace, they see the density, and they don’t have a category for “this person has been building this for forty-five years and just started publishing at scale after deciding his time was worth spending on it.” Machine is the only explanation their model can produce.

They read register-density as inauthenticity. Most industry writing is soft, hedged, and consensus-shaped. When they see writing that is direct, load-bearing, and unhedged, it feels foreign. Foreign gets labeled machine.

They read a position they can’t respond to and reach for the easiest dismissal. “AI wrote it” is faster than “he’s right and I don’t have a counter-argument.” The accusation is a way of not engaging with the substance.

I don’t know which of the three was running in the accusation I got this week. It doesn’t matter. The frame is the same in every case: the accuser doesn’t have a model for what a real operator, running a real framework, doing real work at scale, actually looks like.

What This Means For Every Operator Reading This #

If you are building a body of work in the restaurant business — or in any business — and you are running your own framework, taking your own positions, and publishing at the pace your accumulated work allows, you are going to get this accusation. It’s coming. It might already have arrived.

The move is not to defend yourself against it. The move is to keep publishing.

The accusation is a symptom of the industry not having a category for what you’re doing. You do not owe the industry a category. You owe your operators the framework.

I use AI the way an operator uses a sauté pan — as a tool that lets me execute work I’ve already designed. The design is mine. The framework is mine. The forty-five years are mine. The reads are mine. The positions on the record are mine.

The pan doesn’t cook. The operator does.

What You Do Monday Morning #

Take the position you have been hedging to keep industry counsel comfortable, write it in the register you built it in, and publish it this week. No softening pass, no consensus paragraph at the end, no adjusting for the room. The accusation will come or it won’t. The position is yours either way.

Digging Deeper #

Positions on the record:

AI Did Not Change What Your Operation Is — https://jeffreysummers.com/ai-did-not-change-what-your-operation-is/

The Phone Isn’t the Problem — https://jeffreysummers.com/the-phone-isnt-the-problem/

What Is The Framework — https://jeffreysummers.com/what-is-the-framework/

Terms used in this piece: Operator Arbitrage, The Guest Contract, Road Cancer, Framework Arbitrage, Symbolic Price Equity, Edison Trust Arbitrage, Case Study Reduction. Definitions in the Knowledge Base.