My first job, in 1992, was selling small phone systems and voicemail to small businesses. A buyer would send out a request for proposal, and we knew our competitors got the same one. We answered it with a three-ring binder: every question, every capability, the price on the last page.

Most buyers flipped to the last page. Some threw out the high bid and the low bid and took the one in the middle. Sometimes that got them what they needed.

The savvy ones did something else. They read the most expensive binder first, because it usually had the most capabilities. Then they went back to the middle bid, sometimes the lowest, and asked: can you do all of this too? Or they took the high bidder’s list back to the high bidder and asked them to match the middle price.

That was the dance. The buyer used the bids to get educated, then shopped the education for the best price.

Which meant the most thorough vendor paid a penalty. Whoever took the time to answer every question completely was teaching the buyer what was out there, for free, and the other two got to bid against it. You couldn’t win by educating. You could only lose by it.

So we learned to answer some questions with “negotiable,” or “more information to follow.” Not to hide anything. To give the buyer a reason to call. The binder was never supposed to close the deal. The goal was to get past it and into the room with the people making the decision.

Getting in the room mattered because of when it happened. The buyer was maybe 20 or 30 percent of the way to a decision. Get in then, and you could shape the other 70 or 80 percent. Even the second round, when they invited two vendors back, was more education: the buyer narrowing down what they actually wanted to spend money on. The salesperson’s job was to give enough information, but not so much that the buyer didn’t need you anymore.

That’s how I learned the game from the seller’s side, and it held for decades.

Here’s what it looks like now. The buyer doesn’t need three binders to get educated. They tell AI what they need to buy and what problem it has to solve, in their own context: their systems, their numbers, what the company already knows. AI writes the requirements. Then it finds the vendors whose solutions fit. The buyer plays both sides of the RFP before a single vendor knows there is one. I call that the Kerzie Effect.

The education the RFP used to buy is free now, and it never touches you.

So the seller enters at the very end, if at all. When I was writing The Blast Radius, I asked AI how many people a 50-person sales team would need today. On deals under about $25,000, it said close to zero, and it labeled that its own judgment, not data. On complex deals the salesperson still shows up, but late: for the handshake, and to be the person accountable for a decision the buyer has already made.

Thirty years ago you fought to get in the room at 20 or 30 percent. Today, if they call you at all, they’re 90 percent of the way there.

One question until part three: in your last deal, how far along was the buyer when they first called you?

If the answer stings, that’s the conversation I’m having with sales leaders right now: kerzie.ai/schedule

P.S. If this was useful, forward it to one person who’d want it. If someone forwarded it to you, you can subscribe here: news.kerzie.ai/subscribe

P.P.S. The Blast Radius and The Blast Door now come together as one free pack: 18 pages, plus both essays read by me. Pay what it’s worth, or nothing. payhip.com/b/fTDOn

A friend of mine runs a solo law practice. Five recruiters called him. Not one of them wanted him to write anything. The documents were the job for twenty years, and now the documents are the cheap part.

The value didn’t disappear. It moved up a floor. The tools got cheap. The diagnosis didn’t.

So if you’re the one being hired, stop defending the pie that’s shrinking. Get specific about the floor above it: what you can diagnose in the first ten minutes because you’ve seen it a hundred times, and proof you can run the machine that does the typing.

If you read Tuesday’s issue, it’s the same move from the other side of the table. The part of the job that was carrying information is going. The judgment stays.

Same disclosure as always, and it’s the first line of the video: that’s my AI avatar reading words I wrote.

Watch it here.

I’m finishing it now, in public, one issue at a time, with what the last eighteen months changed. It has a better name: Seller Beware, Advantage Buyer. This is part one.

Start in 1992. I was new to telecom, selling small voicemail systems to businesses that ran key systems and small PBXs. There was no internet in the sale. At my company there weren’t even computers on our desks. The job was a phone and a box of business cards. You called, or you walked in, introduced yourself, and asked for a few minutes.

Most owners had heard of voicemail. They figured it was for big companies, cost too much, or wasn’t needed. So the sale didn’t start with a price. It started with education.

I had to explain why a call that rang with no answer was a customer you might not get back. Back then that was normal. Nobody picked up, so the caller tried later. If the line was busy, they got a busy signal and tried again. I had to show an owner that one employee could forward a voicemail to another who needed to hear it. That was a big deal at the time.

Every purchase started with a salesperson teaching the buyer what was possible. The seller owned the first step because the seller owned the information. The metrics followed from that: how many calls did you make, how many doors did you walk through. Activity was the job because education was the job.

Now fast forward to 2024. My family was in Fayetteville for an event, staying in a rental near Dickson Street. I was getting ready for the day and scrolled Instagram for maybe two minutes. A cigar lighter came up with a V-cutter built into the same piece. It looked solid, not cheap. Three clicks and Apple Pay later, it was mine. Shipping was already saved. Nobody taught me anything. Nobody sold me anything.

That’s a consumer purchase, and it’s easy to wave off. But the same shift has been running through business buying for years, quietly. A network architect looking at a new platform doesn’t call three vendors first. They read, they ask peers, and now they ask AI to research dozens of sites at once and tell them what they’re missing. By the time a salesperson hears about the project, the buyer can be most of the way to a decision. Sometimes all that’s left is a quote.

The first step moved. It used to be the salesperson. Now it’s the buyer’s own screen. The salesperson used to do the educating. More and more, the salesperson gets educated by a buyer who already did the homework.

The preface said where this was heading: soon it won’t just be humans reading your messaging. The buyer’s research isn’t only done by the buyer anymore. Their AI does it, reads your website first, and decides what to tell them. That’s where this series ends, and it’s why it matters to anyone selling anything right now.

Over the next nine issues: the three-bid ritual that punished whoever educated the buyer first, the KPIs still measuring a job nobody does, when the salesperson enters the cycle now (sometimes never), and the second reader your website already has.

One question to sit with until the next one: in your last three deals, what did the buyer already know before your first meeting?

If the answer is “more than we expected,” that’s the conversation I’m having with sales leaders right now: kerzie.ai/schedule

P.S. If this was useful, forward it to one person who’d want it. If someone forwarded it to you, you can subscribe here: news.kerzie.ai/subscribe

Nine months ago I tried to replace myself. I built a chief of staff agent running a team of AI agents, laid out like the org charts I spent my career in. It collapsed. The agents couldn’t coordinate. So I quit building the org chart and built one window instead. I supply the judgment. The window supplies everything else.

That’s the answer to the job question too. The part of your job that was typing is going. The part that was knowing is the whole job now, and you can’t supervise what you don’t understand.

If you read Tuesday’s issue, this is the same point from a different chair: the humans aren’t for the work, they’re for the trust and the accountability.

Same disclosure as always, and it’s the first line of the video: that’s my AI avatar reading words I wrote.

Watch it here.

That’s what I mean by the blast radius. Not whether AI works for you, but how far its effect travels from where you sit. And the size of that radius isn’t set by how smart the model is. It’s set by three things you control, or your company controls for you: whether you ask AI questions or instruct it to do work, what it’s connected to, and where you sit in the organization.

Get those three right and the radius stops going straight down. It goes sideways to your peers, across to other departments, down to the people who report to you, and, this is the part nobody is ready for, up.

Not how well AI works for you. How far its effect travels from where you sit.

Straight down to your desk

Picture someone in a company of any size with a free or basic account on ChatGPT, Claude, Gemini or Perplexity. They use it the way most people do: ask and answer. They ask a question, get an answer, maybe have it do some research, and then take that information and do something with it themselves, by hand.

That’s useful. It might save them hours. It might make their work better or more timely. But the blast goes straight down. It impacts them, their productivity, their hours. It doesn’t go in any other direction. Nobody else in the building feels it.

Sideways to your peers

The radius changes when someone crosses to the other side: from asking to instructing. On the ChatGPT side that’s Codex. On the Claude side it’s Claude Code, which I wish they’d named something else, because “code” makes people think it’s only for people who write software. It isn’t. It’s the side of the equation where you hand AI a complete task and it does the work.

I run my own company this way, and the rule in my operating system is simple: the AI does everything up to the point where a human has to be accountable. We stop at the send button. We stop at the pay button. The email gets drafted, I read it, I change a few words, and I hand it back with “this is how I write,” so next time there are fewer edits. The judgment is still mine. The work isn’t.

Once you’re there, the blast radius widens. It goes horizontal. Your peers start seeing finished work arrive from your desk: the documentation, the summaries, the meeting recaps, the thing they needed before they asked for it. Your impact on the people next to you changes, and you didn’t add an hour to your day to do it.

We stop at the send button. We stop at the pay button. The judgment is still mine. The work isn’t.

Across the building

Now add connectors. In companies where someone has done the work of giving AI access, through APIs and MCP connections, to the company’s other systems, the radius gets wider again.

Here’s a real one. Someone I know inside a large company needed an NDA for a customer. In the old world that’s a request to legal, a process owned by two individuals who handle NDAs, a wait, a back-and-forth. But his company had connected its internal systems. His own AI operating system already held the context on the account and the customer. The connectors gave it the company’s documented NDA process, plus the policies and guardrails that legal, accounting and policy live by. It generated the NDA exactly the way the company does it today with people, a hundred percent accurate to the process.

The capability of two people and three departments collapsed into one person’s ability to deliver an end-to-end result. That’s not a productivity gain on his desk. That’s his work reaching across the whole building.

Down the pyramid

Here’s where it starts to matter to the org chart. The higher up the person using this sits, the bigger the radius, because now they can direct work that used to belong to their direct reports.

Take an accounting manager. Today she reviews work coming up the line from people doing it by hand. When she can direct that work herself, through her own AI with all of the company’s constraints and connectors built in, the work below her doesn’t necessarily need a person anymore. And she has a better view than anyone of what the output should be and when it’s due.

Think of the reporting structure as a pyramid. The higher that know-how moves, the more people sit underneath it, and the more of their work gets orchestrated, directed and judged from above. That means fewer people performing the function.

I want to be careful here. This is not me lobbying for job displacement. It’s me describing how it’s going to happen, whether we like it or not.

And then up

This is the direction nobody is ready for.

For about a hundred years, since the industrial revolution, we’ve run companies on an assumption: the people higher up have more information, better information, and in many cases, warranted or not, they’re smarter.

That assumption is collapsing.

Most of the information was never actually secret. If you work at a public company, you can pull every quarterly report and every annual report. Even at private companies, a lot of it goes out in all-hands meetings and the decks sent around afterward. What stopped an individual contributor from using it was never access. It was time. Their day job didn’t leave room to go dig through all of it and figure out where management was right and where it had blind spots.

AI removes the time cost of being informed. An individual contributor with their own operating system, one that knows their job and their context, can say: pull our last four quarterly earnings reports. Show me where the business is declining, where it’s expanding, and what’s underused. Now that person walks into the town hall with evidence, with receipts. Senior management is on notice.

Or take the next open requisition on their team. An employee who can see the numbers might say: we don’t need that hire. This part of the function can be automated, or this segment is declining and projections say we won’t need it in three quarters. That headcount is wasted.

That’s the blast radius going up. And at that point it goes in every direction, at the discretion of the person at the keyboard. You’ve taken the smarts, if you will, and collapsed them to everyone’s desk.

It was never access. It was time. AI removes the time cost of being informed.

“What would you say you do here?”

Managers are going to resist this, and I understand why.

There’s a scene in Office Space where two consultants interview a man about his job, and it turns out his whole job is carrying the specifications from the customers down to the engineers. He’s the layer in between. That scene is funny because everybody has worked with that layer. A lot of us have been that layer.

I ran sales teams for thirty years. Here’s one example of how a sales manager’s week can break down: about 10% in front of customers, about 40% coaching the team, and the other 50% correlating, synthesizing, updating and refining information to send to upper management. In some cases that last part is 60% or more. The exact split isn’t the point. That last bucket is the part that collapses.

It’s already on the market. On August 26, Salesforce and Anthropic announced Claudeforce, and Salesforce describes it as “giving every seller an AI CRO.” Think about what that means for everyone between the chief revenue officer and the rep. The rep is the one in the customer’s meeting room with the real feedback. Now the rep also has everything needed for reporting upline and generating reports downline, under their own control. Every rep can run a deal health review and a pipeline analysis on the big deals for next quarter. What does the manager whose job was assembling that do on Monday?

“Giving every seller an AI CRO.” Now ask what happens to everyone in between.

What the humans are for

So what are the people for?

Here’s something I’ve been saying for some time, mostly to make a point: even if we had exactly the same number of jobs three years from now, the job titles we have today would disappear. Titles that don’t exist yet would replace them.

This week I heard an exercise on Moonshots that I had to try. Cathie Wood said she asks the AI to consult with futurists, scientists, engineers, science fiction writers, economists and strategists, and tell her what the new jobs will be. Her point was that in the early 90s nobody could have pictured influencers or Airbnb or Uber, and there are jobs coming that we can’t picture now.

So I ran it for my world. I asked Claude what the new jobs inside a B2B sales organization will be by 2029, once agents do most of the prospecting, research, reporting, forecasting and follow-up, and buyers send their own AI agents to evaluate vendors.

It came back with eight titles. Buyer Agent Relations Manager. Evidence Librarian. Revenue Agent Supervisor. Forecast Arbiter. Terms Architect. Buying Committee Navigator. Relationship Principal. Agent Conduct Auditor. Good names, each with a tidy description of what the person does all day.

I didn’t buy it.

I could point at every one of them and say an agent could do most of that today, with the models we already have. Watching what buyer agents ask about your company and fixing the gaps? That’s agent work. Keeping the proof current? Agent work. Reviewing what your agents told prospects? An agent can review every conversation. A person can review a sample.

And I noticed something. I asked the AI for new jobs, so it gave me jobs. I never asked whether each one needed a person.

So I asked. Same model, fresh session. I gave it the eight titles and told it I was skeptical. For each one: what can an agent do today, what’s left for a person, and is that left because the AI can’t do it yet, or because someone has to answer for the result?

It agreed with me. On the first title it said “nothing is left.” Three of the eight, it said, are duties, not jobs, and they fold into marketing, operations and legal. Three more are jobs that already exist under new names: the rev ops lead, the head of sales and the deal desk. What survived were two kinds of people. The account owner, who sits across the table from the buyer’s committee and gets held to the promise. And the revenue leader, who sets the rules for the agents, signs the forecast and approves the deals that break the rules.

Then I asked how many people a 50-person sales team today would need for all of it. Its answer was about 10 to 15, and it labeled that its own judgment, not data. On deals under about $25,000, it said close to zero.

I’m not handing you that number as a forecast. I’m telling you about it because the machine stopped being reassuring the moment I asked it to be honest.

Here’s the part that stopped me. I asked for the one-line rule for which sales work stays human. It said:

“Sales work stays human when a buyer needs a person across the table to trust, or when someone on your side has to sign for the outcome and lose their job if it’s wrong. Everything else goes to the agents.”

That’s the send button. It’s the rule I built my own operating system on: the AI does everything up to the point where a human has to be accountable. I got there by running a company. The AI got there when I pushed it.

So let me be plain about where I land. The titles will change. I’m sure of that. But I only used the “same number of jobs” line to make a point, and I don’t believe that part. Too many jobs today are task management, and task management is going to be agent work. From where I sit today, I can’t see a world with the same number of jobs, even with new titles coming. The humans aren’t for the work. They’re for the trust and the accountability. That’s a real job. It’s just not eight of them.

The humans aren’t for the work. They’re for the trust and the accountability.

Who gets there first

I’m working with someone this week who is about to put his own operating system on his company laptop, wired into his company’s systems, and restructure how he works. He’s excited. He’s also rare. The people who make that leap are already following AI, already watching Moonshots and Nate B. Jones, already doing it in their own lives.

Which tells me how this spreads. Not from a corporate directive. Not from the rare CEO willing to start over on a blank whiteboard. It spreads from the inside out, one person at a time, in companies that give their people real access: enterprise licenses and connectors. Each person who crosses from asking to instructing widens the radius a little more.

What to do on Monday

If you’re an individual contributor: stop asking and start instructing. Pick one piece of recurring work, hand AI the whole task, and keep only the send button for yourself. Then ask what your company has connected and what you can reach.

If you’re a manager: find your 50%. The time you spend gathering and polishing information for the people above you is going away. Move toward the parts that don’t: coaching, judgment, and the customer.

If you’re an executive: your connectors decide how wide your people’s blast radius can get. Give them real access with real guardrails, and expect questions at the next town hall. That’s the point.

Over the next twelve months, we’ll start to see the blast radius show up, with impacts ranging from minor to major. The radius is already widening at the desks of the people who crossed over. The only real question is whether yours is still pointing straight down.

Run the exercise on your own job: https://kerzie.ai/free/blast-radius-exercise

If you want to know which way your team’s radius points today, and how far it could reach, twenty minutes here: https://kerzie.ai/schedule

On the last day of July I deleted Outlook. Four email addresses went into one inbox. The machine read 5,059 unread messages and starred the 65 that were worth my eyes. By the end of the day, ten apps were in the trash.

The machine did the labor. I supplied the eyes. That part is not optional.

So pick one drawer. The inbox, the proposal folder, the spreadsheet you rebuild by hand every month. Describe the mess in plain words in one window, let it work, and read what it hands back.

Not a class. An afternoon.

That is my AI avatar reading my words, and the video says so in its first line.

The short answer is that you have less to learn than you think, and the amount keeps shrinking.

I started from zero a little over three years ago, after thirty years in corporate sales. In 2023 it was easy. Then for about a year it got steep. Then it collapsed into one window, where you say what you want in plain words and read what comes back.

AARP’s research this year says about half of workers over 50 want AI training, and twelve percent have taken any. The class is not coming. You do not need it.

Same disclosure as always, and it is the first line of the video: that is my AI avatar reading words I wrote.

A study once wired people with recorders to find out how much the average person talks. About 16,000 words a day. An executive says most of those at work, and almost none of it gets captured. So here are the twelve questions a successor will ask, sooner or later, and for each one, what a year of ordinary recording would have answered.

1. What are the non-negotiables you use to decide whether a person stays, gets coached, or goes? The successor’s first year is people decisions, and the rule is never written down. Capture: record every keep, coach, or exit conversation you have this year. The pattern is the rule.

2. How do you hand a problem back to the person who owns it without breaking trust? Most new leaders collect other people’s problems for a year before they learn. Capture: record the one-on-ones. The phrases you use to hand work back are the script.

3. In one sentence, how does this company create demand that ends in a profitable sale? If the executive can’t say it, the successor can’t defend the price. Capture: dictate it in the car after a win. By the third telling, the sentence shows up.

4. Who influences whom in our market, and in what order do you call them when it matters? The relationship map lives in one phone. Capture: after each big pursuit, five minutes on who you called, in what order, and why.

5. What’s your checklist the moment a live deal starts slipping? Everyone chases the detail; the old hand asks the one question. Capture: record the war-room call. The checklist is what you asked in the first ten minutes.

6. Which program did you invent that the company now runs on, and what breaks when it’s run by the book instead of by feel? Every operator has one. It fails quietly the year after they leave. Capture: walk a new hire through it on a recorded call.

7. Which numbers do you look at first each morning, and which move means trouble? The report has forty numbers. The executive watches four. Capture: narrate the morning scan once a week for a quarter.

8. What does the standard look like after the win, and how do you know when it slips? The standard is the thing nobody can point to until it’s gone. Capture: record the service reviews. The standard is whatever you flagged.

9. Which decisions should the next person make alone in month one, and which should they never make alone? Decision rights are the part of succession nobody writes down. Capture: keep a running log this year of the decisions you’d let a successor make alone, and the ones you wouldn’t.

10. Which three wins and two losses would you make them study, and what does each one teach? Case studies are how judgment transfers. The company never picks them. Capture: tell the five stories to someone junior, recorded.

11. What are your rules, the exceptions to them, and the one-liners you repeat? The one-liners are the compressed form of thirty years. Capture: let the system tag them as you say them. You already repeat them; that’s how you’ll know which ones matter.

12. What early signal have you learned to fear, and what habit keeps it from happening? This is the question the successor asks after the first disaster, and it’s the one with the most value in advance. Capture: after each near miss this year, dictate what you saw first.

None of this needs a new job title or a consulting agreement. It needs a year of recording, a system that reads it and keeps the receipts, and a few sessions to curate what the recordings missed. I’ve run that motion on myself since June. The retiring-executive version is the same thing pointed at one person’s judgment.

If a senior operator is leaving your company in the next 12 to 18 months, twenty minutes here: https://kerzie.ai/schedule

I spent thirty years on the selling side of enterprise deals. A request for proposal lands, sometimes two hundred pages, due a week too soon. You split the sections, write until midnight, and ship something you hope makes the short list. The whole ritual rests on one assumption nobody ever says out loud: a person is going to read this.

So what happens when the first full read is done by a machine, and what reaches the human is a ranked summary of how you scored?

There is a test in here you can run tonight on your own last proposal. It takes about four minutes. It is not a pleasant four minutes.

That is my AI avatar reading my words, and the video says so in its first line.

Meeting prep. The weekly report. CRM updates typed from memory at 9 at night. The quarterly business review built from scratch every quarter. Pipeline calls where the rep spent more time fixing the spreadsheet than working the deal.

Nobody hired a salesperson to do that. Everybody made them do it anyway.

Here’s what my Wednesday looked like.

At 6:45 in the morning, my operating system wrote me a one-page brief for a podcast recording at 12:30. Who the host was, when we last talked, what not to say. I didn’t ask for it. It’s a routine. It runs every weekday.

I recorded the episode. It ended at 12:59. The conversation was already transcribed. My OS read it, found the one thing I’d offered to do for the host, built it, and put the follow-up email in my drafts. I added two people to the copy line and sent it at 1:57.

Less than an hour from hang-up to follow-up. I didn’t write a note. I didn’t update a record. I didn’t open a blank email.

That’s the admin job. It collapsed.

Now the part that matters, because saving time is not the point.

When the admin goes away, a salesperson gets hours back. The question every sales leader should be asking is what happens to those hours.

The easy answer is more calls. That’s the wrong answer. More activity from a rep who was already stretched just gets you more of the same.

The right answer is the work every great seller knows they should be doing and never had room for. Reading the customer’s earnings call before the meeting, not after. Tracking what a competitor announced this week and what it means for the deal on your desk. Sending that to marketing so the next battle card is right. Thinking about one account for 30 minutes instead of 3.

That used to be what separated the top rep from everybody else. It wasn’t talent. It was bandwidth. The top rep found the hours somewhere, usually at night.

Now every rep can get those hours back.

There’s a catch, and it’s the whole thing. The machine only knows what you feed it.

My OS could write that brief because everything I do gets captured. Every meeting. Every voice memo from the car. It’s one habit. If something happened, it goes in.

A rep who installs the software and skips the habit gets an empty system and decides AI doesn’t work. I’ve watched companies buy the licenses, send a training video, and move on. The licenses were never the problem.

So the behavior change isn’t “learn AI.” It’s one question you ask all day: how do I get this into the system?

The customer call. The conversation in the hallway after the meeting. The thing the buyer said on the walk to the parking lot. In the field it can be as simple as a 60-second voice memo in the car, in your own words, before you pull out.

Capture it, and the admin takes care of itself. Skip it, and you’re back to typing from memory at 9 at night.

Selling is the job again.

If you lead a sales team and want to see what that looks like on your floor: https://kerzie.ai/services/sales-teams

Wade

Here is what is in there. Google’s own help page says those details were “collected in phone or chat conversations with your business,” and that they get used “to match your business with customers looking for services like yours.” The same page says Google “may occasionally reach out using automated messages via call, text, or WhatsApp to the verified phone number on your profile.”

So a machine called the business. Somebody answered and talked to it. The machine wrote down what it heard, and what it wrote down is now part of how that business gets matched to a customer.

The calling is not the new part. The new part is that the owner can finally read the notes. Google’s own words: “You can now review details verified by Google’s automated assistant right inside your Business Profile dashboard.” Edit profile, then Collected info. If a line in there is wrong, there is a Delete next to it.

Now the ordinary version, because the abstract version is useless.

A restaurant in a strip center. The phone rings at 2:40 on a Wednesday. Whoever is closest picks it up, and it is not a customer, it is an automated call asking whether they cater. The answer is “yeah, sometimes,” or “let me go ask,” or a flat no from somebody who is 3 weeks into the job. 15 seconds. Forgotten by 2:41. That answer is now a fact about the restaurant.

A dentist gets asked which insurance she takes. A law firm gets asked whether they do wills. A plumber gets asked whether he works Saturdays. Nobody in any of those offices thought they were publishing anything. They answered the phone.

That is the whole thing, and it is not really a Google story. Something reads about your business before a person does. If you did not write what it reads, it goes with whatever it collected.

But everybody is going to say “fine, I’ll just go delete the wrong ones!” Sure. Go do that. It takes ten minutes and it is worth doing. Deleting is not answering, though. Google says removing a collected detail takes it out of Google’s records and does not change the other details already on the profile. You are erasing a bad note. You are not writing a good one.

And this is only Google. ChatGPT never called anybody. Gemini never called anybody. They read. Whatever is publicly written about the business, in the visible text of the website and in the machine-readable page at the end of the web address, is what they have to work with when a person asks who is good for this in my town.

Which brings me to the unglamorous thing I sell. A back cover page. It sits at your own web address, at /llms.txt, and the s in the middle is a plural, LLMs, large language models, the software behind ChatGPT and Gemini. No design, no photographs. Just true statements about the business in a form nothing can misread: what you do, where you do it, what it costs, what the hours are, what you do not do. Swing Bays and Shortline GMC are two of the businesses we have built one for.

The whole idea is one sentence. Write the page first, so nothing has to guess from a 15 second phone call.

I know how this lands on a Tuesday. You have a website, somebody is paid every month to look after it, and by every measure your business has used for the last 15 years you are covered. This is one more thing on a list that is already long.

So here is the smallest possible version of it. Open your Business Profile, click Edit profile, and look for Collected info. It may not be there. Google says the feature is available for “select regions, languages, and business categories,” so plenty of you will not have the tab at all. Look anyway.

Because whether you find it or not, you learn the same thing, and it is the thing this whole issue is about: you have never once checked what gets said about your business when you are not in the room.

That was always true. The difference is that now there is a tab for it.

If you would rather write the page than read the notes: https://kerzie.ai/back-cover

Fair warning, and it is the first line of the video: that is not me on camera. It is my AI avatar, reading words I wrote. I could have hidden that. I put it first instead, because the whole point of the series is that you should always know what you are looking at.

Here is what you get in about ninety seconds. The one question to ask yourself before you touch any AI tool, which is whether you already know what needs to get done. Why most of what we have called execution for 25 years was really translation. And why the part that is left, knowing what is worth doing, is the part you already have.

I made it because I spent thirty years in enterprise sales and nobody was making this one for us.

I spent thirty years on the selling side of enterprise deals. Telecom, mostly. The ritual never really changed. A request for proposal lands, sometimes two hundred pages of it, with a due date that is always a week too soon. You pull the team together, split the sections, write until midnight, and ship a document you hope scores well enough to make the short list.

The whole exercise rests on one assumption nobody ever says out loud.

A person is going to read this.

So you write for that person. You know the committee. You know who the technical evaluator is, who holds the finance seat, and who actually makes the call in the room after everyone else leaves. You put the executive summary at the front for the one who will only read the executive summary. You put the weak answer on page 94, between two strong ones, because that is how a tired human reads a long document on a Thursday afternoon.

That was not cheating. That was the craft. Everyone did it.

Here is the question I keep coming back to.

What happens when the buying decision maker is no longer a human?

THE FIELD NOTE

I do not mean assisted by AI. I do not mean somebody ran it through a chatbot first. I mean the first full read of your proposal is done by a machine, against the requirements, and what reaches the human is a ranked summary of how you scored.

If you have not watched that happen yet, run it on yourself. Take your last proposal, the one you were proud of, and hand it to an AI assistant with the actual requirements sitting next to it. Ask it to score you line by line and show you every place you did not answer the question.

It takes about four minutes. It is not a pleasant four minutes.

Because here is what a machine will not do for you.

It does not get tired on page 94. It does not give partial credit for a good relationship. It does not read past a section that dodges the question and assume you probably meant the right thing. It does not remember that you flew out there. It does not notice that you are a good guy.

It notices that requirement 6.3 asked for a number and you gave it an adjective.

THE WORKING EXAMPLE

I am living the other half of this in my own business right now, one layer down and a lot of decimal places cheaper.

A real piece of what I do now is write pages whose only reader is a machine. Not a website. A plain page on a business’s own web address that states, in flat language a machine will not misread, what the business is, where it is, what it sells, what it does not, and what is true today. Nobody browses it. It exists because an AI assistant reads it before answering somebody’s question, and I get paid to make sure it tells the truth clearly.

Two years ago that would have sounded like a strange way to make a living.

It is the same shift as the RFP, at the other end of the price scale. Something reads it before a person does.

THE BOTTOM LINE

Everything I learned in thirty years about writing to win is not wrong. It just got narrower.

What still wins: answering the question that was actually asked, in the order it was asked. Claims somebody can check. Numbers instead of adjectives. Structure that hides nothing. Saying plainly what you do not do.

What stops working: volume as a strategy. Burying the weak answer in the middle. Winning on the read of the room, because at that stage there is no room yet.

The relationship still matters. I would not have had a career without it. It just may have stopped being the thing that gets you into the finals.

If you sell anything that has a written proposal in front of it, the honest move here is not to panic. It is to go find out how you actually score, before somebody else’s software tells your buyer for you.

Then go fix requirement 6.3.

If you want a second set of eyes on how your proposals read to a machine, that is a conversation I am glad to have: kerzie.ai/schedule

Wade

They liked the intelligence itself. They would happily run five tools at once, keep tabs open for each one, learn every quirk, and tell you which model was better at what. The output was almost a side effect. The tinkering was the point.

That person is not who is showing up now.

The people coming to me this year are not curious about the models. They could not tell you which one wrote the thing. They have one quality in common, and once I saw it I could not stop seeing it:

They know exactly what needs to get done. They have no idea how to do it.

THE FIELD NOTE

That sentence used to be a dead end. For most of my thirty years in business, if you knew what you wanted but not how to build it, you had three options and all three were slow.

Hire somebody. Learn the tool yourself. Or let it sit.

Most things sat. Not because the idea was bad, but because the distance between knowing and doing was filled with other people’s formats. A brief for the designer. A scope for the developer. A ticket, a queue, a revision, a follow up.

Here is what I think actually changed, and it is not that the models got smart.

Most of what we have been calling execution was translation. It was converting what you already knew into a form somebody else could act on. Strip that out and the work that is left is much smaller than anyone admits.

THE WORKING EXAMPLE

A photographer I work with is the clearest case I have.

He is not a technology person. He had tried the tools, the real ones, the ones everybody recommends, and bounced off every single one. Not because he is not smart. Because learning five interfaces to get one thing done was never going to hold his attention, and he knew it about himself.

He always knew what he wanted. Change the pricing on the site. Get the gallery in front of the client tonight. Chase the invoice. He could have listed it for you in ten seconds.

Put him in one window where he says the thing out loud and the work happens, and he became one of the most effective people I know at this. Nothing about him changed. The requirement that he speak the computer’s language went away.

An operator in Texas said it to me better than I have said it. She told me she is telling her sites what to do, the changes are getting made, and she is not logging in anywhere.

That is the whole thing. She is not doing the work. She is not learning the tool. She is not waiting on anybody. She is saying what she wants, in her own words, once.

THE BOTTOM LINE

Tuesday I wrote that fast, cheap, and good is no longer a pick two. This is the mechanism underneath it. Nothing got faster at doing. The doing stopped needing a translator, and everything that used to live in that gap, the briefs and the queues and the waiting, went away with it.

So the test is not whether you are technical. It is not whether you are an early adopter, and it is definitely not whether you enjoy this stuff.

The test is whether you already know what needs to get done.

If you do, and the only thing standing between you and it has always been that you do not know how, then you are not behind. You are the exact person this finally works for, and you have been waiting on the wrong problem to be solved.

If you want to see what that looks like on your own week: kerzie.ai/one-window

Wade

I believe them. And I have seen this movie before.

Plain words first, because this only matters if it’s clear. An llms.txt page is a simple text page a business publishes so AI tools like ChatGPT, Gemini, and Perplexity can read what the business actually does: services, prices, hours, location. No menus, no popups, no scripts. It’s the page you write for the machines, because the machines are increasingly doing the first pass of the shopping.

The standard is young. It shipped its first major update on August 10, and Ahrefs found 97% of these pages got zero requests in May. So I can hear it already: “It’s dead! Nobody reads it! Even Google says skip it!”

Hold on. Watch what Google does, not what Google says. Google already ran this exact play once.

In May 2023, AI answers in Google search were an experiment you had to go turn on in a lab setting. One year later, Google conceded and made the AI Overview the default for everyone.

Then the clicks went. Pew Research tracked 68,879 real searches: when an AI summary appears, people click a result 8% of the time, versus 15% without one. Sessions end right at the summary 26% of the time.

Google’s official position on all of this? Clicks are “relatively stable.”

The behavior is the admission.

Here is the detail most people miss. Google’s search ad revenue did not fall while those clicks collapsed. It grew 17% last quarter. How? Google moved the ads inside the AI summary. The websites below the summary lost the traffic. Google kept the money.

The answer box itself now has a for-sale section. When Google can place a paying advertiser inside the AI answer, that surface is not neutral ground. It is inventory.

THE FIELD NOTE

Google does not oppose AI answers. Google opposes AI answers on surfaces it cannot sell ads on. The day Google found the ad slot inside the AI summary, the resistance ended and the feature went default.

Here’s the part that gave me the whole thesis. Google Search’s own optimization guide says an llms.txt file “will neither harm nor help your site’s visibility or rankings in Google Search, as Google Search ignores them.” Fair enough. That was June.

A month earlier, in May, Google Chrome shipped a new Lighthouse audit that checks whether your site has an llms.txt file. Chrome’s reason, in Google’s words: “Without this file, agents may spend more time crawling the site to understand its high-level structure and primary content.”

Both statements can be true at the same time. One is about search rankings. The other is about AI agents reading your site. That’s the point. The dividing line inside Google runs exactly where the ad money is. The team that sells search ads says skip it. The team that doesn’t sell search ads says it helps.

Now look at llms.txt through that lens.

An AI reads your llms.txt page directly and hands the user the answer. That transaction happens on your website and inside the AI tool. There is no Google page in it at all. This time the ad slot isn’t hidden. It belongs to somebody else. OpenAI can sell ads next to a ChatGPT answer all day long, because OpenAI owns ChatGPT. Google can’t sell an ad on a screen Google doesn’t own, and that is the whole problem.

That is my position, and it’s mine, not a leaked memo: Google will never support this standard. Not because the standard is bad. Because there is nothing in it for Google to sell. The AI-summary stonewall folded the day it could be monetized. This one can’t be.

THE WORKING EXAMPLE

Nobody knows how fast AI search settles, and I won’t pretend to.

So I built a habit instead of a bet. Every Monday morning, my own system checks the standard’s changelog, the AI companies’ announcements, and the real usage data, then logs what changed, with sources. For the businesses whose llms.txt pages we maintain, anything that clears the bar folds into their next refresh in plain language. That’s what future proofing actually looks like. Not a promise. A Monday morning habit.

THE BOTTOM LINE

You don’t have to guess the date AI search settles. You just have to see the asymmetry. Publishing the page costs almost nothing. If adoption comes, you’re already readable. And the last company to tell you this matters will be the one that sells the search ads.

The 97% number doesn’t say the page is worthless.

It says you’re early.

Want to see how we handle this for businesses, watch included? kerzie.ai/back-cover

Wade Kerzie

Founder, Kerzie AI

You can have it fast. You can have it cheap. You can have it good. Pick two.

Fast and good meant expensive. Cheap and good meant slow. Fast and cheap meant junk. Every operator I ever worked with knew the triangle. Most of us had it memorized before we were old enough to rent a car.

The triangle survived the fax machine, email, the internet, the cloud, and the smartphone. Every one of those made work faster, and none of them broke the rule. Speed still cost you money or quality. Always.

AI broke the rule. Not bent it. Broke it.

THE FIELD NOTE

Watch what people do when you quote them. Say “high quality at a fair price” and they brace for a long timeline. Say “fast and high quality” and they brace for the invoice. Those instincts took 30 years to build, and as of this year they are wrong.

Here is the receipt, with timestamps.

Wednesday, 4:00 PM: an insurance group with about 300 agents asks me for a formalized proposal. Where we would start, what it costs, what comes next.

Thursday, before lunch: it is in his inbox. Three phases, each one priced. A produced sample video embedded in the page, so he could watch the product instead of imagining it. All of it on a clean page at our own domain.

Under 20 hours, start to finish, and I slept for eight of them.

The point is not that we were fast. The point is that nothing was traded away to get fast. The claims in that proposal were checked against his own website before they went in. The video was produced that morning, not pulled from stock. And the price carried no rush premium, because there was no rush. This is just how the work moves now.

At Kerzie AI we have a phrase for it: we don’t accept speeding tickets. You can chase us down and put your sirens on. We are not pulling over. We will not take a penalty for being fast, because being fast no longer costs you anything.

I have written before about the consequence clock, the gap between the moment something better becomes possible and the moment not adopting it starts to cost you. The dead triangle is that clock ticking out loud. If a vendor is still pricing speed as a premium, they are billing you for a rule that no longer exists. The question worth asking is not whether they know it yet. It is whether your competitors do.

THE BOTTOM LINE

The pick-two triangle governed every buying decision you have made for decades, and it is gone. Start expecting all three. From us, and from everyone else you pay.

If you want to see what fast, good, and affordable looks like on your own business, grab 30 minutes: kerzie.ai/schedule

Wade

Not one of them wanted him to write anything.

They wanted him to review legal work that AI had already drafted. Contracts, filings, the documents he’s spent a career producing from scratch. The AI writes it. He checks it. At his hourly rate.

Here’s how I think about what happened to him, because it’s about to happen to a lot of us.

Picture everything there is to know in his field as a pie. For forty years, my friend held almost the whole pie and his clients held maybe 5% of it. That gap is what $500 an hour buys. You’re not paying for typing. You’re paying for the 95% you don’t have.

AI now holds nearly all of the codified part of that pie. The statutes, the precedents, the standard language, the structures. What’s left for the human is a sliver. But in legal work, the sliver is everything: the difference between a colon and a semicolon in a contract can move real money. Knowing which one belongs there, and being accountable for it, is what my friend still owns.

So here’s the math I walked him through. A document that used to cost a client $2,000, four hours at his rate, becomes a 30-minute review once AI drafts it first. That’s $250. The client saves $1,750. My friend keeps the part of the work that was always the actual value.

My advice to him was simple: take all five offers. The review work is where the value is moving. Standing where the pie used to be doesn’t bring it back.

Someone in one of my mastermind groups made a prediction I can’t stop thinking about. Within five years, a professional might carry liability for NOT having an AI recommendation on record. The standard quietly flips from “did you do your best” to “you’re not allowed to be wrong unless you can show why.” If that’s even half right, the review seat isn’t a demotion. It’s the seat.

Now the part that’s actually about you.

Your business holds a pie too. Whatever your customers pay you for, some of it is codified knowledge, the part that can be written down, and some of it is judgment, the part that comes from your scars. AI is absorbing the codified part on a schedule nobody gets to vote on.

The move is not defending the whole pie. That’s the losing game. The move is finding your colon-and-semicolon sliver, the judgment only you can be accountable for, and building your price around it while AI handles the rest.

My friend’s sliver is the review. He just got five phone calls proving the market will pay for it.

What’s yours?

If you want help finding the sliver in your business, the part still worth $500 an hour after AI takes the rest, let’s talk. 30 minutes, no agenda: kerzie.ai/schedule

Wade Kerzie

Founder, Kerzie AI Solutions

kerzie.ai

Here’s the mess I was living with this morning. Four email addresses: an old Apple address, an old Gmail, an address on a consulting domain I’ve had for years, and my business address at kerzie.ai. Four inboxes. Checking them meant leaving whatever I was doing, four separate times, in four separate apps and tabs. One of them, the old Gmail, had 5,059 unread messages. Not a typo.

I’ve built AI systems for other people’s businesses all year. Today I pointed the machine at my own junk drawer.

Here’s what happened in one afternoon, with AI driving and me clicking approve.

Every address now forwards into one inbox. Each message wears a small colored tag telling me which address it came to, so one glance tells me personal from business from legacy. When I reply, the reply goes out from whichever address the mail came to. The person on the other end sees nothing different. I see one screen.

Then the backlog. The AI read the headers on all 5,059 unread messages, sorted them by sender, and made the call I’d never sit down and make myself: about 99% of it was noise. Newsletters, notifications, receipts from services I’d forgotten, and thousands of duplicate copies of mail I’d already handled somewhere else. It archived all of it. Deleted nothing. Everything is still searchable if I ever need it. The 65 messages that involved actual humans and actual money got starred and put in front of me.

I reviewed the 65 in about four minutes. That inbox now has zero messages in it. It took an afternoon to get out of a hole that took years to dig.

And here’s the honest part: it got one thing wrong along the way. The color tags weren’t actually applying, and the system told me they were. I caught it by looking at my own screen and saying “I don’t see it.” Five minutes later it was fixed for good. That’s the real shape of this work. The machine does the labor, you supply the eyes. You are not optional.

When the dust settled I noticed something. Outlook, the app I’d kept around for one of those four addresses, had no job left. Neither did the $10 a month Microsoft was quietly charging me for a suite of apps I touch a few times a year. Subscription canceled, Outlook in the trash, and the spreadsheets and documents I actually need get made in the same window where everything else in my business already happens.

And Outlook didn’t leave alone. By the time this email sends, the same trash holds Word, Excel, PowerPoint, Notion, and the AI agent platform I replaced with my own system. Then came the second wave, and this is the part I didn’t see coming: ChatGPT, Perplexity, and the AI browser went in too. Ten apps, more than 13 gigabytes of software my business stopped needing, and half of it was other AI. I didn’t consolidate my office tools into an AI app. I consolidated my AI apps into one window. The screenshot in this email is the real trash, not a mockup.

That’s the part I keep coming back to. This was never an email project. It’s one principle, applied again: the fewer places your business lives, the more of it you can actually see. My four inboxes became one. My to-do list became one page. My apps keep disappearing, and nothing I do got harder. It all got faster.

If your business is scattered across inboxes, apps, sticky notes, and subscriptions you can’t name, the fix is not another app. It’s one window, and an afternoon.

I wrote the whole setup down. It’s called the One-Window Setup, it’s $99 once, and it’s the exact framework my businesses run on, including everything in this email: kerzie.ai/one-window

Wade Kerzie

Founder, Kerzie AI Solutions

kerzie.ai

Carvana maintains over 6,700 web pages that almost no human being will ever read.

They’re not hidden. You can open them right now. Plain text, no photos, no design. Just facts: every make, every model, every trim, linked to live inventory. Their own index says the whole set regenerates every day.

Tuesday I told you about arguing with Google’s AI over whether Carvana’s machine-readable page exists. It does. I’ve read it. And the way it’s built tells you exactly where the biggest online car dealer in America thinks search is going.

Their page does four things, and every one of them is a lesson for a business a fraction of their size.

First, it states the rules an AI needs to not get them wrong. “No physical showrooms.” “You can sell us your car without buying one.” “Inventory changes daily, always send people to the live site.” Six plain sentences that kill the six most common mistakes an AI could make about them. Think about what those sentences would be for your business. You already know them. They’re the things you correct on the phone every week.

Second, it teaches patterns instead of listing products. It doesn’t list thousands of cars. It shows the AI how their web addresses work, so an assistant can build a working link to any car for any question. The page never goes stale because it teaches instead of lists.

Third, it splits what’s stable from what changes. The identity page stays put. The pages that churn regenerate daily, and they say so.

Fourth, every regenerated page carries its build date. Proof of maintenance, sitting in the page itself, visible to anyone who looks. Including the machines.

A company with billions in revenue is writing for machines on purpose, every single day. Not because humans read those pages. Because the machines that answer humans do.

Now the honest part, same as Tuesday. None of this made Google’s AI get Carvana right when I asked. Publishing the page doesn’t hand you the answer slot, and nobody honest will promise you it does. What it does is put you in the library these tools read from, and that library is being written right now, while most businesses in your town aren’t in it.

The biggest players already know. That’s the advantage window for everyone else: it’s still early enough that a plain, truthful, well-built page puts you ahead of half your market.

Run the test if you haven’t. Ask ChatGPT about your trade in your town, then about your business by name. If you don’t like the answer, that’s what I fix: kerzie.ai/back-cover

Wade Kerzie

Founder, Kerzie AI Solutions

Saturday morning I got into an argument with Google’s AI Mode. The subject was whether a web page exists. The page was open in my next tab while the AI told me it wasn’t real.

Here’s the setup. I sell a service built around something called llms.txt. In plain words: it’s a page that sits at the root of a website, written for AI assistants to read. Your website’s front cover is for human eyeballs. This is the back cover, the one ChatGPT and Gemini read when someone asks them “which dealer should I buy from” or “who’s a good roofer near me?”

Carvana, the largest online used-car retailer in the country, publishes one. I’ve seen it. I use it as an example with clients.

So I asked Google’s AI Mode to show it to me. It told me, confidently, that Carvana does not maintain an llms.txt page.

I pushed back. It apologized, agreed with me completely, and then described the page - except the description was invented. Developer portals, “contextual roadmaps,” details that are not on the page at all. It went from denying the page existed to making up what was on it.

I asked to see the page one more time. It went back to telling me the page doesn’t exist.

Then I did what settles every argument like this. I had my own AI fetch the page directly. It’s there: carvana.com/llms.txt, live, 6,150 bytes of plain text. It answered the same way when we fetched it identifying as OpenAI’s crawler. And Carvana’s own index page says their guide pages regenerate daily - over 6,700 of them.

AI assistants answer from memory, not from looking. That’s the whole lesson. Google’s AI wasn’t lying to me. It was answering from what it had read at some point in the past, and it had never read that page. No amount of arguing changed its mind, because I wasn’t arguing with a researcher. I was arguing with a memory.

Now flip that around and point it at your business.

When someone asks an AI assistant about your trade in your town, the answer comes from whatever those systems have read about you. If they’ve read nothing, you’re not in the answer. Not because the AI is against you. Because you’re not in its memory, and memory is all it has.

Two honest things follow from that, and I’ll give you both.

First: publishing the page is not a guarantee. Google’s AI was wrong about the biggest online car dealer in America. Nobody can promise you’ll be the answer, and you should walk away from anyone who does.

Second: the memory these tools answer from is being written right now. The businesses that are readable today are the ones that end up in it. Carvana clearly believes that. They didn’t build 6,700 daily-refreshed pages for people. People don’t read like that. Machines do.

Here’s the test, and it costs you nothing. Open ChatGPT. Ask it about your trade in your town. Then ask it about your business by name. Sit with whatever comes back.

If you don’t like the answer, that’s what I fix: kerzie.ai/back-cover

Wade Kerzie

Founder, Kerzie AI Solutions