Are You Next?

What two founders learned by sharing skills, testing AI, and building around the work only humans can do

Part One: The Connection, the Audit, and the Operating Plan

---

A copywriter named Devon offered to audit my store. On paper that's a simple exchange — he studies weatheredsailor.com, finds where the words lose people, and hands back the changes he believes would make the site clearer and more persuasive. That's his service, and it's real work.

But his offer set off a question I wasn't entirely comfortable asking: how much of what he does could the system I've been building already do?

Not because I wanted him to be replaceable. Because I've spent the last few years trying to understand a technology that has a lot of people in knowledge work quietly wondering whether they are. Some of the work I'm building now sits in exactly that category — SEO and answer-engine strategy, helping companies stay legible to search engines and to the AI assistants increasingly standing between people and what they're looking for. Which means I'm learning to sell and perform the exact kind of knowledge work everyone keeps telling us the machines are coming for.

That makes for a strange relationship with the technology. Some days it's hard not to read AI as part of what unsettled the career I'd built. Other days it's just as clear that learning to build with it is how I make whatever comes next. I could stay angry at it — some days I am — but anger doesn't teach me where it's useful, where it's dangerous, where it's mostly theater, and where a person still matters so much that removing them makes the work worse. The only way I know to answer those questions is to point it at something real.

Weathered Sailor is that real thing. Not a demo. Not a clean case study built so the machinery looks smart. A small business with actual customers, thin margins, emotional weight, technical debt, incomplete data, and consequences when something breaks.

So when the copywriter offered his audit, I didn't want to quietly feed his report to a model and see whether it could imitate him. I wanted to propose a trade.

How two strangers ended up here

The part of this I keep coming back to may have nothing to do with the technology.

We didn't meet because I went looking for a consultant. No marketplace profile, no star rating, no cold pitch. Weathered Sailor sponsored a retreat, a conversation happened between two strangers on a social network, and somewhere in it two people who build things for a living recognized each other. One — Devon, who runs a copywriting studio called Wordsmithery — has built his craft around how words move people. The other had spent the last few years building a strange little operating system of models, agents, data connections, and approval gates around a clothing company. That slowly became a question: could we trade capabilities instead of invoices?

I've cold-hired people online before. Some were genuinely good. But the structure has a gap in it. The consultant is responsible for delivering the recommendation; the owner is responsible for living with whatever happens after it touches the business. If it's hard to implement, or fights the economics, or bends the brand the wrong way, those consequences don't always land on both people equally. That's my experience, not an indictment of every consultant.

What this has instead is harder to buy: a person with an actual stake in the outcome. He's invested. I'm invested. Neither of us wants to be the one who wasted the other's trust. That's a kind of accountability a review score can't quite hold.

What he's giving me

His side of the trade is the service he sells: a full copy-and-conversion audit of the store — what it communicates now, where the offer and headlines and calls to action fall short, where the voice is strongest, where proof is thin — and then the changes he'd make.

He gave me a structured scorecard across the major elements of persuasive copy. Brand voice emerged as the clear strength — the conclusion I'd defend hardest myself. The largest weaknesses were the first impression and the trust layer: the homepage hook, credibility, and social proof. Calls to action, value-proposition clarity, audience resonance, and search visibility all needed work as well. His clearest conclusion was the one I agreed with most: the brand already has a distinctive voice, but that voice isn't doing enough work in the exact places where a stranger decides whether to understand, trust, or buy.

I don't read those grades as laboratory measurements. They're an expert's structured judgment — a visible theory of where the store is strong, where it's weak, and where his work could move it. That gives us a starting hypothesis we can test without publishing the proprietary scoring behind his service. We'll keep his original assessment privately as one baseline, then report publicly on whether the identified areas improved and what happened to the business measures underneath them.

What I'm giving him

My side is stranger.

I'm taking his analysis and his eventual copy and absorbing them into the tooling I've built around Weathered Sailor — not to ask a model to rewrite his sentences, but to see whether the system can understand his reasoning, challenge it without flattening it, verify the assumptions under it, connect it to the live storefront and analytics and product economics, turn it into controlled implementation, make or stage the changes safely, and measure what actually moves. And then find the places where all of that still can't equal the judgment, taste, and restraint of a person who's spent years learning the difference between copy that functions and copy that connects.

The point isn't to produce a winner. It's to map the work. The experiment has at least three honest outcomes. The system executes some of it worse than he can. It reaches parity where the work is structured and repeatable. And it outperforms either of us alone where it can inspect more data, hold more state, and catch inconsistencies across systems that a person simply wouldn't sweep by hand. Then there's the part it can't finish correctly without him — or without me. That boundary is the whole thing I want to map. Not to shrink his value to whatever the systems cannot reach, but because identifying where his value is most uniquely human may help him protect it, strengthen it, and build more of his business around it.

And there's a second half I care about more than the scoreboard. If a real chunk of the surrounding work — extraction, research, consistency checks, implementation planning, measurement setup, QA, reporting — can be absorbed into tooling, then maybe he spends far more of his hours on the work he actually got into this for: understanding the customer under the demographic, finding the sentence that isn't just competent but true, knowing when a claim persuades and when it manipulates. I once put "five times more time" on it as an example. The multiple isn't the point; the leverage is. If it works, he walks away with more than a report about the threat — he leaves with his own human-and-AI capability map, run on his own service, plus tools, access, and the practical skill to keep experimenting on his own. A clearer moat around the human work only he can do.

Two businesses, one experiment between us. He helps me see where Weathered Sailor fails to say what I think it is. I help him see how the technology changes the mechanics — and the defensible human value — of what he does.

What happened when his audit entered the system

I gave the work its initial direction, answered a few consequential questions, and let it run overnight.

By morning his scorecard wasn't sitting by itself. It was one of three independent lenses, deliberately set against each other. His audit held the professional copy view. ChatGPT ran a separate challenge — was the positioning coherent, were we treating subjective scores like measured outcomes, did the store even have enough traffic for the experiments we were sketching, were we optimizing for revenue when margin was the more honest number. And Reggie — my COO-style orchestrating agent — and the specialist agents under him examined the live operation: the storefront, the templates, the catalog, Shopify, Klaviyo, GA4, product costs, guarantee economics, defects, claims risk, and whether the words still sound like me.

I built it so they wouldn't agree by default. Averaging three opinions into one comfortable number doesn't make the answer wiser — it just hides the disagreement. So I made them argue, and made every proposed change trace back to a specific row in the tracker. If an action couldn't be tied to a row, the rule was to stop and flag it, not improvise. Not glamorous. Also one of the clearest lines between asking AI for ideas and letting it near the operation of a real business.

The most convincing moment wasn't when the system produced something impressive. It was when it corrected itself.

An early pass called a blank "You saved" label a sitewide defect. The verification agent checked all 287 products and found it firing on exactly zero of them — latent dead code, not a live wound. A second issue looked like a recommendation-engine bug, product pages suggesting themselves back to the shopper; the full sweep found the real cause was 34 duplicate "twin" products in the catalog, a governance problem wearing a code problem's clothes — and because some of those pairs could be legitimate, the system refused a blind bulk cleanup and flagged each one for a human decision. Then the email number: earlier analysis had it converting around six percent, dramatically above the rest of the store. Flattering. Under the light it was two orders across thirty-three sessions, riding on an analytics pipeline that was itself broken and capturing only about a quarter of the store's traffic. The verdict wasn't "email doesn't work." It was promising but under-sampled — the opportunity kept, the seductive number kept from hardening into a fact. And ChatGPT pushed back on measuring success by revenue per session at all: the store is largely made to order, a returned item can't always be resold, so the plan moved to contribution margin per session as the more truthful north star. It was right. I changed the plan.

That's the kind of thing I hoped the checks and balances would catch — the moments when I, or the expert, or the technology are each tempted to believe a convenient story.

What the audit became

This is the part that vanishes when I say "I ran it through my tooling."

The scorecard didn't become a slightly longer scorecard. It became a project. By morning the original assessment had turned into a connected set of operating artifacts: a full copy-and-conversion audit of the live store; a reconciled master plan; a phased roadmap; a tracker covering decisions, assumptions, measurement, email evidence, guarantee economics, and analytics repair; five written decision records; a phase-zero verification brief; and an execution handoff assigning every downstream task to a specific agent. Human approval stayed mandatory before anything could publish, send, reprice, delete products, or change a customer promise.

The diagnosis was bigger than "the copy needs work," too. The real problem was orchestration — the brand already has the hard-to-fake ingredients (a real founder story, a real mission, collection writing with its own voice, deep product specs, free U.S. shipping, a stated resolution promise) but they don't consistently show up in the right order, close enough to the moment a shopper needs them. A customer doesn't experience copy, UX, catalog, trust, and merchandising as separate disciplines. They experience one storefront. So the roadmap got organized around a single continuous argument — this is for someone like me; I get what it is; I see why it's different; I believe the quality; I trust the company; I understand the wait and the return policy; other people validate it; I know what to do next; it feels worth the risk — and every change now has to strengthen one link in that chain or fall out of scope.

That's what his scorecard became inside the system. Not a replacement for his thinking. A structure built to carry his thinking much farther into the business. An operating system, not a hunch.

Where the system stops

Here's the honest status: almost none of the customer-facing copy is live yet. The plan is built; the proof is still ahead.

The verification work has largely closed. Shopify is the commercial source of truth for now, with GA4 untrusted for revenue until it's repaired. The email claim has been properly downgraded. The margin definition is set. The defects are scoped more accurately than my first draft had them. What's still ahead is the actual experiment — and the part no model gets to decide.

Which idea should the homepage lead with? The product? The feeling of having weathered something? The quality? The mission? The nautical metaphor? What stays literal and what stays poetic? What makes a stranger understand the brand in three seconds without sanding off the thing that makes it ours? A model can generate the routes. Agents can critique them. Analytics can tell us what happened after. None of them has the authority to decide what Weathered Sailor is willing to become in pursuit of a conversion. That's still mine. And the actual sentences still have to survive the person whose whole profession is knowing when technically correct words don't yet feel alive.

That's where the copywriter's direct judgment becomes most important — not as a reviewer bolted on after the system finishes, but as the person whose expertise the entire system is trying to understand, preserve, and amplify. The system can do a remarkable amount around that expertise. Whether it can carry it without distorting it — and exactly where it can't — is the experiment.

Why I'd rather do this together

It would have been easy to make this adversarial — take his report, run it privately, and post about whether AI beat a copywriter. More clickable, probably. It also misses the part I find most hopeful about the whole thing.

A small act of sponsorship made a connection. The connection made trust. The trust made a trade neither of us could have bought off a service menu. Now two founders are letting each other inspect the machinery behind their businesses. Neither of us knows how it turns out — which is what makes it an experiment and not marketing. Maybe the tools prove more capable than he expects. Maybe his work exposes where my system is shallow or overconfident or has no taste. Most likely both. But I'd rather map that boundary next to someone with a real stake in the answer than keep talking about AI as a distant thing that just happens to people.

Maybe that's the larger thing I'm still learning. The technology doesn't have to enter every relationship as a replacement. Sometimes it can enter as the thing two people study together. Sometimes the person worried about being displaced learns to build with it, and the expert examining what the technology means for his own service comes out with more leverage instead of less value. And the best thing that came out of "networking" here wasn't a customer — it was two strangers realizing they each held something the other needed, and deciding to make both businesses stronger by sharing it.

What comes next

This is Part One — the connection, the trade, the original scorecard, the three competing lenses, the overnight system, and the checks that corrected our first conclusions.

Part Two starts when the recommendations become live, customer-facing work. We'll preserve his original scorecard privately as one starting reference, then pair its directional findings with the measures a scorecard can't hold: first-screen comprehension, product-page engagement, add-to-cart behavior, conversion, contribution margin per session, average order value, return and replacement cost, email performance, search visibility, customer feedback — and whether the language still sounds like Weathered Sailor after every layer of technology has touched it. Some changes will be obviously right and need no test. Some we'll release in sequence against a preserved baseline. A few high-exposure ones might justify a real split test. Others will need actual customer conversations, because pretending a small store has enough traffic for statistical certainty would only manufacture confidence.

Then we compare what happened against what he expected. Which parts of his diagnosis held up? Which conclusions changed once they met the operating reality? Where did the technology carry his thinking faithfully, and where did it lose something important? What got faster, cheaper, more scalable — and what still needed a person who'd spent years learning to feel the difference?

I don't know where that line falls yet. That's the point. But after watching a technology reshape the kind of work I've done for most of my life, there's something meaningful about using it this way — not to prove another person is unnecessary, but to help both of us understand what we can build now, what we can do better together, and what remains worth protecting.

I'm starting to think the strongest moat isn't hiding what you know. It's understanding which part of it is uniquely yours — and then collaborating in ways that let you spend more of your life doing exactly that.

That's what I mean by Are you next? Not next to be replaced. Next to find the people, tools, and opportunities that help you discover which part of your work is most human — and build more of your future around it.

We started as two strangers with different capabilities. Now each of us gets to help the other find out what his business might become.

---

1% of every order supports To Write Love On Her Arms, Active Minds, and NAMI.