A seller asked my AI to write his emails

Last summer, an experienced enterprise seller found my AI clone and asked whether it could write his prospecting emails.

When I checked the logs, I wanted to scream.

Not because AI can't write emails. It absolutely can. It was because he was about to delete the only real advantage he had left.

Being himself.

I wrote a post about it. The point: AI trained on generic data gives you back generic garbage. AI trained on your data gives you your genius at scale.

I believed it (and I still do).

Here's the part I left out.

I trained the models on my own writing. Thousands of posts, seven years of articles, lessons, and essays, and my own frameworks. I shaped every output with my perspective. By my own rules, I was doing it right.

Growth still happened, but something was off.

Then I found the words for it: I had stopped writing and started reviewing. The loop had changed. I was editing something a machine produced instead of struggling to produce something myself.

I missed the old ritual of struggling. The good kind. The professional craft tax of writing 2,500 bad words before 1,200 worthwhile ones could emerge.

So I installed Pangram (no affiliation), which flags whether a post is human-written or machine-generated, and looked at my own feed.

There were tons of red AI indicators on most of my posts (ironically, including the one about the enterprise seller using my AI to write emails for him).

I'm not deleting them. I consider them a record of the detour.

Then I spent six months heads-down working (and struggling) on my book and remembered what I'd been missing.

It wasn't that I used AI. I made life-changing income selling conversational AI to the world's largest brands. This essay is orchestrated with the help of AI. I use AI every single day. I’m bullish.

What I got wrong was the order.

I'd been treating my body of work like a mine—something complete, that you can keep taking from.

But it’s more like a garden. You can harvest a garden forever, but only if you keep planting new seeds and nourishing the soil.

If you stop planting, the yield drops slowly enough to not notice, because what's already there still looks fine.

And the machine can't restock it for you. It can build plenty of impressive things, but it can't create your body of work, because yours is made of things that actually happened to you. What it produces is generated from patterns, not from personal principles or lived experience.

When you let the AI create without your input, you stop being the person who made the thing worth pointing at in the first place.

What changed with my approach wasn't the tools. It's who holds the standard. I do the thinking. I decide what's worth saying. I bring the scars, the receipts, and the point of view. Then I spar with a machine to sharpen it, and I throw out everything that doesn't sound like me.

That is not the same as asking it to be me.

It isn't less work, either. It's just different work. What I can't hand over is the taste.

I sketched this on my whiteboard this week, and it's the simplest way I can explain what I mean.

For years, I’ve been preaching “Human First. Seller Second.” There’s room for a third layer now with AI.

Human first. What we all want is pretty basic and universal—to be seen, to be missed if we're gone, to be understood, and to be of use. What we need doesn't change either—physical, mental, emotional, relational, and spiritual health. The Human part of us is what fuels our battery. Nothing runs without it.

Let me illustrate this point.

I'm an introvert with social anxiety. Crowded rooms drain me. Networking events leave me craving a dark room and a cozy blanket. However, deep work, building, tinkering, and writing … these give it back.

When I started in sales, everything I heard suggested I needed to be extroverted, high-energy, and charismatic. “Fake it till you make it.” Always “be on.” I tried. But the more I tried to be someone else, the more depleted I got, and the worse I sold. I was trying to be a seller first and ignoring what made me human.

Seller second. If you’ve completed the Autonomy Assessment, you understand that strong professional performance is built on four ideal components: independent thinking, tranquil conditions, protecting your well-being (including having real psychological safety in your role), and consistently making an impact.

That last one is where we measure our craft on a continuum:

Chaotic → Baseline → Pro → Elite → World-Class

Go deep in one area of genuine interest and real expertise, and you move along that line. And if you're stubborn enough and lucky enough, maybe, just maybe, you eventually get to world-class in one capability. But if you stay scattered, you sit still.

AI third. It’s last on purpose.

If you start with AI, then you skip the part where you become worth listening to. If I open a blank prompt every morning instead of writing, I never get from deficient to baseline, then pro, then elite at writing. And then I have nothing worth feeding the machine.

“Accept your initial agitation.”

After I pay that “tax,” I ask this one question before I point AI at anything:

Does this serve and enhance my human and professional wants and needs?

If yes, aim it and let it run. If no, you're outsourcing the best parts to a machine.

Keep the structure in this order and the path to improving reveals itself.

Here's a map I came across recently that reframed how I think about my own AI setup.

I've seen a few versions of this floating around the internet over the past month. Here’s what matters for us.

There have been essentially four eras of working with AI systems since ChatGPT came along at the end of 2022.

Each era absorbed the one before it. The newest rage is called graph engineering.

Greg Isenberg explains it simply: prompt engineering is about asking the AI a better question. Context engineering is about giving it better information. Graph engineering is how you design the work around the AI, so the whole thing stops living inside one messy giant chat.

Knowing more was the old game. Orchestrating better is the new one.

Almost every seller I talk to is still in era one, maybe two. They try to cram everything into one giant chat and haven’t set up projects with instructions and skills. They give a prompt, get one bad answer, and end up with low-quality outputs. And most of them tell me they feel behind.

I get it. The rules change every six months.

But they're not behind on AI. They're behind on the internet.

It’s been around our entire careers, and most sellers I know have never once used it to share what they actually know. Not a single post, not one breakdown of a deal they ran. The best distribution ever built, sitting there unused.

AI doesn't close that gap for you. If you've never written anything down, there's nothing for the machine to point at.

You can climb every rung of this ladder and still end up with a very fast way to sound like everyone else.

More agents don't mean better work. Sometimes it just means more noise—five AI workers confidently repeating the same wrong idea. And a lot of AI work fails for a simpler reason than people think: the same model that writes the answer also grades the answer. Greg said it well in the video: “It's like asking someone to write their own performance review and then being shocked when they describe themselves as a visionary.”

That's why the human must create the event and check the work. The evidence has to come from outside the system. This is your own judgment about what "better" means.

You don't need to learn to code to move up this ladder. You need to stop asking for outputs and start designing a system—what goes in, who does what, what gets checked, and where you step in.

My graph starts with voice notes and whiteboard photos, not prompts. I have separate workers set up to absorb sources, draft, and cut anything that doesn't sound like me. Then a detector I don't control grades the result, and I decide what to cut.

A while back, I asked, Which 10% is yours?

A lot of you replied. I read every one.

Here's what I'd do with that answer now.

First: you have more of a body of work than you think. Every pre-call email. Every deal review. Every account narrative you wrote before a first meeting. It's scattered across multiple drives, various apps, numerous folders, and three employers. You've never once looked at it as an asset.

It is the asset.

Second: mine is writing. Yours probably isn't.

Yours might be finding the problem worth solving. Building trust with a stranger who controls the budget. Turning something complicated into a narrative an executive forwards to their board. Knowing three months early which deals are real and which are a dead-end.

Whatever it is, that's where the system points. Not at your whole job. At that one thing.

Aim it there, and use it to turn that skill into a capability—something you can summon on demand, at a high standard, under pressure, again and again.

A skill is something you have. A capability is something you own.

That's the difference between leverage you're leasing and leverage you keep.

One more thing, and I should have said it a long time ago.

I write from the far side of this. That's a problem.

I know how it reads when someone tells you to protect your calendar while you're at 5% of your quota in month seven of a new role, still learning who's who, watching the ramp window shut behind you.

I wasn't always here. In my first year in strategic sales, I didn’t close my first deals until month 12 and grossed just over $200K. The million-dollar years came after that. And before any of it, when I was just starting in sales, I was $35K in debt, leaving NYC to start over and sell $45 ads for a local events paper in Sarasota, Florida.

The system I'm describing doesn't require you to be free already.

It's what you run while you're not.

My book has a new working title:

Permanent Leverage: How Tech Sellers Turn Peak Earning Years into a Life No Employer Can Touch.

My editor told me the old one, Autonomy Architecture, wasn't grabbing. I sat with it for a day, didn't want him to be right, but he was. The new name is mine, not his, and he hasn't weighed in on it yet. We talk in a few hours. He may well talk me out of it.

Then I searched the manuscript and found something worse than a weak title. The word "leverage" appears 34 times across 57,000 words, and I never define it once.

Whatever the name ends up being, that's the real problem. So here's the definition, and it's going in the opening pages:

A lease and leverage are the same idea pointed in opposite directions.

And Learning Path 6 is getting more focused by design.

It’s built with one promise:

In 90 days, you'll set up your AI system and point it at the most valuable skill in your arsenal, to turn it into a capability that builds permanent leverage.

The knowledge will be free. What I'll charge for is the practice.

It isn't scarce anymore—you can get the framework from the same AI I use. What's scarce is someone looking at yours and telling you the part you got wrong.

Stay human, friends.