Ship Something That Pays

Core concept 01

Describe it, and watch it get built.

On February 2, 2025, Andrej Karpathy, formerly Tesla's AI director and a founding OpenAI engineer, posted a tweet that named the whole trick: "there's a new kind of coding I call 'vibe coding,' where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." He meant describing what you want in plain language and letting the model write it, instead of typing every line yourself. The term stuck hard enough that Collins Dictionary named "vibe coding" its 2025 Word of the Year.

Here is the honest part most ads skip. Simon Willison, a longtime software engineer, drew the real line: if the AI wrote it and you reviewed it, tested it, and can explain how it works, that is software development. If you can't, and you ship it anyway, that's vibe coding, and his rule is to never commit code to a real project he couldn't explain to someone else. Good territory for it: a landing page, a first-draft deck, a personal tool nobody's business depends on. Bad territory: anything touching money, passwords, or a stranger's data without a second pair of eyes. Inside that honest fence, the tools are genuinely good. Claude Artifacts turns a description in a chat into a real working page or small app, and since July 13, 2026 you can publish it to a public link anyone can open, no coding and no Claude account required to view it. Gamma does the same trick for decks, documents, and full websites from one prompt. This lesson is about using that honestly.

Sources: Andrej Karpathy, Feb 2025 · Simon Willison · CNN, on Collins Word of the Year · Anthropic, on public artifacts

Core concept 02

The 2.9% is the smallest number on the page.

Everyone worries about the payment fee. It is the least important line in the whole model. Stripe charges 2.9% + 30¢ per successful charge with no monthly fee, and Payment Links are included at no extra cost. Gumroad charges 10% + $0.50 on a direct sale, also no subscription. On a $150 workshop seat, Stripe keeps about $4.65 and you net roughly $145. That fee is not what decides whether the offer works.

What decides it is the line nobody puts on the page: your delivery time. A digital guide costs you nothing to hand to the next buyer, so its second sale is nearly pure margin. A live workshop costs you the hours in the room every single time, so its economics look completely different even at the same price. Same $150, two different businesses. The processor fee is rounding error next to that difference, which is exactly why fee-shopping is the wrong thing to optimize and offer shape is the right one.

Go deeper: gross margin, and why it is the real number

Gross margin is what is left after the cost of delivering one more unit: price − processor fee − your marginal cost to fulfill it. For a digital download the marginal cost is essentially zero, so margin is roughly 90% on Gumroad, 97% on Stripe. For anything you deliver with your own hours, the honest marginal cost includes those hours at a rate you would actually accept, and the margin drops accordingly.

Why this matters more than the sticker price: a $47 download at 95% margin and a $300 service at 40% margin can net the same dollars, but only one of them scales while you sleep. Knowing which business you are building is the decision. The price tag alone hides it.

Sources: Stripe, pricing · Gumroad, pricing

Scroll the diagram sideways →

Same price, two businesses · the processor fee is the sliver; delivery time is the story

Core concept 03

Price the outcome, not the hour it took you.

The instinct is to price from the inside: add up your time, tack on a margin, post the number. Buyers do not price that way. They pay for the outcome and the alternative they are comparing you to, and both of those are usually worth far more than your cost to produce. A wedding photographer's gallery-delivery template is twenty minutes of your afternoon; to the photographer drowning in delivery emails, it is a Saturday back.

This is why the same deliverable can honestly carry a $27 price or a $270 price depending on who it is for and what it saves them. The number is a positioning decision, not an accounting one. Set it against the outcome and the next-best option the buyer already has, then check that the math still clears your floor.

Go deeper: anchoring, tiers, and the middle option

Price is read in context, never in isolation. A single $150 option is just a number; a $90 / $150 / $290 ladder makes the $150 look like the sensible middle, which is why so many offers ship with three tiers even when the top tier rarely sells. The top tier is not there to sell. It is there to make the middle feel chosen.

Use it honestly: real differences between the tiers, no fake decoy nobody could want. The lever is legitimate context, not manufactured pressure, which keeps it inside the "you're not late" ethos rather than the countdown-timer one.

Source: Paul Graham, "Do Things That Don't Scale"

Core concept 04

Offer shape sets the ceiling before you write a word.

There are three shapes a first paid thing can take, and they cap your income in different places. A one-time sale, a workshop or a download, is the easiest to ship and the easiest to understand, but every month starts back at zero. A productized service, a fixed-scope offer at a fixed price, pays the most per sale and builds reputation fastest, but it is still your hours. A recurring offer, a membership or a subscription, is the hardest to start and the only one where last month's work still pays you this month.

None of the three is the right answer. The right answer is to know which one you are building, because it decides everything downstream: what you price, how many buyers you need, and whether "sold out" is a good day or a trap. The person who ships a one-time workshop thinking it will become passive income is not wrong to ship it. They are wrong about the shape, and the shape is knowable in advance.

Go deeper: the ladder most small offers actually climb

The common path is not one shape, it is a sequence. A one-time workshop proves people will pay for the topic at all. The recording plus worksheets becomes a download, the same content sold on repeat with near-zero delivery cost. The people who want more become a small recurring cohort. Each rung reuses the last rung's asset, so the second product is cheaper to build than the first.

This is why "ship the small one-time thing" is not settling. It is the cheapest possible test of whether the whole ladder is worth climbing, run for the price of one afternoon.

Scroll the diagram sideways →

Offer shape · the ceiling is set before you build the page, not after

Core concept 05

Ship the simplest version, then run the two-week test.

Anthropic's advice to its own developers, the same line from Lesson 01, is the discipline here: start with the simplest thing that works. Paul Graham told founders the same thing in 2013, before any of this was AI-assisted, "do things that don't scale," solve the problem by hand for as long as you can. A one-page site with a payment link is the by-hand version, and it is the correct choice, not the lesser one. Engineers are not paid for building the complicated version. They are paid for knowing when the complicated version is a waste.

The reason to keep the build tiny is that the build was never the test. The test is whether anyone pays, and you want to reach that answer for the price of one afternoon, not one quarter. So ship the small version and give it two honest weeks in front of twenty real people. Then read the result like an operator: nobody clicked is a reach problem, plenty clicked and nobody paid is an offer problem, a few paid is a signal to make the next rung. Each of those points at a different fix, and none of them is "build something fancier."

Sources: Anthropic, "Building Effective Agents" · Paul Graham, "Do Things That Don't Scale"

The number you leave with

Your one-page unit economics, before you build anything

Fill these six lines for the offer you are actually considering. It takes about four minutes and it is the difference between a priced offer and a hopeful one. Real numbers only.

01Your pricethe number on the button, e.g. $150
02Minus the processor feeStripe 2.9% + 30¢, or Gumroad 10% + $0.50
03Minus your cost to deliver one more$0 for a download; your honest hourly rate × hours for anything live
04That leaves your margin per saleline 01 − line 02 − line 03
05Your goal for this montha real number, e.g. $1,000
06Buyers you need = goal ÷ marginline 05 ÷ line 04 — if that count feels impossible from 20 people, the price or the shape is what changes, not the effort

Fresh from the lab

What changed this month

Gumroad, one of the two checkout tools this lesson runs on, just started running your inbox for you. On July 2, 2026, Gumroad shipped an Agent tab right inside its app: you type what you want handled, a refund, a payout question, a reply to a buyer, it proposes the action, and you approve it before anything actually happens. It sits next to the Gumroad CLI the company shipped a few months earlier, so a whole store, products, sales, payouts, support, is now workable by a human typing sentences instead of clicking through screens.

A free, mainstream way to have AI build your page just got a lot more free. Canva opened Code 2.0, its AI website and app builder, to every account including the free tier on July 15, 2026, across all 265 million monthly users. Twenty prompts a day at no cost, sixty on paid plans, is plenty to get a first landing page live. It's a real third option next to Claude Artifacts and Gamma for the "have AI build the page" step below.

The AI building your page can now keep working after you close the laptop. Anthropic started rolling Claude Cowork out to mobile and web on July 7, 2026, and extended its doubled usage limits through August 5. Start a build task on Cowork, shut the lid, and it keeps running, so the page in this lesson can come together while you're doing something else entirely.

Sources: Letters by Burk, "Gumroad Shipped a CLI and an AI Agent" · VentureBeat, "Canva launches Code 2.0", July 15, 2026 · TechCrunch, "Anthropic's new Claude feature is quietly selling you on AI", July 9, 2026

Vocabulary

Seven words for the ship-it phase

Four are about building, three are about the economics underneath. Learn all seven and every builder tool, pricing page, and "we just launched" post gets easier to read the right way.

Vibe coding

Describing what you want in plain language and letting AI generate the working thing, coined by Andrej Karpathy in February 2025. Great for drafts and prototypes; get a second set of eyes before anything touching money or data goes live.

Landing page

One page built to get one action: a booking, a purchase, a signup. Not your whole website, not your whole business. Just the page this month's offer needs.

MVP

Minimum viable product. The smallest version worth putting in front of real buyers, built to answer "will anyone pay," not to be the finished thing.

Deploy

Publishing your build so it lives at a real, working address anyone can visit. With today's tools this can be one click, or one share link, not a server you manage.

Gross margin

What is left from a sale after the processor fee and your cost to deliver one more. Near 90% on a download, far lower on anything you deliver with your own hours. The number that says which business you are really in.

Conversion rate

The share of visitors who take the action the page is for. On a small warm launch, a handful of buyers out of twenty is a real signal, not a rounding error. It is how you read the two-week test.

Recurring revenue

Income that renews without a new sale each time: a membership, a subscription. The only offer shape where last month's work still pays you this month, and the hardest to start.

Make it yours

Ship one money page this month

Same concept, your business. Pick your lane and follow the steps. Every workflow below ends in one live page with one way to pay or book, in front of twenty real people. Budget an afternoon, once. A free Claude or Gamma account and a Stripe or Gumroad account are enough to start.

For female founders

I have three offers I keep meaning to put a real page up for, and "this weekend" has been the plan for two months straight.

The pilot-cohort page

Tool: Claude Artifacts or Gamma for the page · Stripe Payment Links for checkout

Pick the thing. Not all three offers, the one you already know works because you've sold it before, even informally.
Gather the raw copy with the assistant from Lesson 01: the price, who it's for, what's included, and three sentences on why it works.
Have AI build the page. Describe it to Claude Artifacts or Gamma and get a real, working one-pager back in minutes, not a wireframe.
Add the payment link. A Stripe Payment Link for the pilot price, capped seats if that's true, no custom checkout build required.
Put it in front of 20 real people. Your list, your last ten client conversations, and the three people who've already asked "do you do this?"

The payoff: the offer that lived in your head becomes a page you can text someone, the same week you decide to ship it.

For artists

My best pieces live in a folder on my phone, and there's nowhere anyone could actually buy one.

The print shop, one page

Tool: Gamma for the page · Gumroad for print and commission checkout

Pick the thing. A print run of three to five pieces, or an open commission slot. Not the whole catalog at once.
Gather the raw copy: sizes, prices, materials, turnaround time, and the plain sentences you'd actually say about each piece.
Have AI build the page. Feed Gamma your images and details and get a real shop page back, formatted and legible, not a grid dump.
Add the payment link. Gumroad checkout per print size, or a Stripe Payment Link for a commission deposit.
Put it in front of 20 real people. Your Instagram close friends list, your studio mailing list, and the people who've already said "I'd buy that."

The payoff: the work that was already finished finally has somewhere to sell, without a single new piece having to get made first.

For actors

I coach other actors through self-tapes between my own auditions, and I have never had one link to send when someone asks how to book me.

The booked-solid coaching page

Tool: Claude Artifacts or Gamma for the page · Calendly for booking, Stripe Payment Links for payment

Pick the thing. One session type: a 45-minute self-tape review, priced once, not a menu of five options nobody will parse.
Gather the raw copy: your credits, who you coach best, what a session includes, and the price you've been quoting in DMs anyway.
Have AI build the page. Describe the session to Claude Artifacts and get a page with your reel, your bio, and the offer laid out clean.
Add the booking and payment link. A Calendly link for the time, a Stripe Payment Link for the fee, both embedded on the page.
Put it in front of 20 real people. Your scene study group, your last acting class, and every actor who's ever asked you to look at their tape.

The payoff: the coaching side hustle stops running through ten separate DM threads and starts running through one link.

For musicians

People ask "do you do weddings?" and "can I buy the record?" constantly, and I still don't have one link that answers either.

The one-link shop

Tool: Gamma for the page · Gumroad for the digital bundle, Stripe Payment Links for booking deposits

Pick the thing. A digital bundle, the record plus two unreleased tracks, priced once. Booking inquiries get their own short form on the same page.
Gather the raw copy: the bundle's track list, three lines on the record's story, and your real booking rate for private events.
Have AI build the page. Describe both halves to Gamma and get one page that holds the shop and the booking pitch without feeling crowded.
Add the payment link. Gumroad for the bundle download, a Stripe Payment Link for booking deposits.
Put it in front of 20 real people. Your mailing list, the fans who comment on every post, and the last five people who DM'd asking about a show.

The payoff: the midnight "do you do weddings?" question gets answered by a link, not by you, at midnight.

For fitness pros

I keep coaching people one-on-one in my DMs for free, because I've never built the page that would let them just buy the program instead.

The 4-week program page

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links or Gumroad for checkout

Pick the thing. The 4-week program you already coach from memory, written down and priced once.
Gather the raw copy: the weekly structure, who it's for, what results look like, and two client wins you can mention without names.
Have AI build the page. Describe the program to Gamma or Claude Artifacts and get a real page back with your program laid out week by week.
Add the payment link. A Stripe Payment Link for a live cohort, or Gumroad if it ships as a self-paced download.
Put it in front of 20 real people. The clients you've already coached this exact program on informally, plus your studio's mailing list.

The payoff: the program you've been giving away in the DMs starts paying you back, at the same quality, for the first time.

For coaches

I run a great workshop live in my head every few months, and it's never once had a real page or a real way to pay for it.

The paid workshop page

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links for the ticket

Pick the thing. One workshop, one date, one price. Not a menu of future topics, just the one you could teach tomorrow.
Gather the raw copy: the outline, who it's for, what they'll walk away with, and the date and time.
Have AI build the page. Describe the workshop to Gamma and get a page with the outline, the date, and a clear price, not a vague "inquire within."
Add the payment link. A Stripe Payment Link for the ticket, capped at the group size you can actually run well.
Put it in front of 20 real people. The clients who've said "I wish you taught this," plus your email list and your last few discovery calls.

The payoff: the workshop stops being a someday idea and becomes a date on the calendar with seats filling.

For therapists

I answer the same three questions on every intake call, and I've never built the page that would answer them without me on the phone.

The self-serve page, with a bright line

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links for checkout · no client details in the builder, ever

Pick the thing first, and the line first too. A downloadable coping-skills guide or a workshop for your niche. Zero client information goes into the AI tool, only your own general practice content.
Gather the raw copy: the FAQ you repeat every intake call, your fees and policies, and the plain, warm language you actually use in session.
Have AI build the page. Describe the resource to Gamma and get a page that answers the repeat questions before the call ever starts.
Add the payment link. A Stripe Payment Link for the guide or workshop, separate from your clinical booking system.
Put it in front of 20 real people. Your waitlist, your Psychology Today profile link, and colleagues who refer to you.

The payoff: the intake hour shrinks because the page already answered the questions that used to eat it, and the red line stayed intact.

For authors

My next book has a real preorder date and zero pages online that actually say so.

The reader-magnet-to-preorder page

Tool: Gamma or Claude Artifacts for the page · your retailer's or publisher's preorder link

Pick the thing. A short reader magnet, a deleted scene or the first chapter, that leads straight into the preorder ask.
Gather the raw copy: the synopsis, the preorder date and link, and the excerpt you're giving away.
Have AI build the page. Describe the funnel to Gamma and get a page that moves a reader from free chapter to preorder in one scroll.
Add the link. Your retailer's or publisher's real preorder link, plus an email capture if you're collecting readers for launch day.
Put it in front of 20 real people. Your newsletter, your last ten reader emails, and the book club that read your first one.

The payoff: preorders start counting before launch week panic sets in, from a page that was live weeks earlier.

For chefs

Private dinner requests come in over text, and I quote every single one from scratch, usually at 11pm.

The private dinner booking page

Tool: Gamma for the page · Stripe Payment Links for the deposit

Pick the thing. One private dinner package, a fixed headcount and price range, not an infinite custom quote every time.
Gather the raw copy: the menu style, price per head, minimum guests, and your real availability.
Have AI build the page. Describe the package to Gamma and get a page with the menu, the price, and the booking flow laid out.
Add the payment link. A Stripe Payment Link for the deposit that locks the date.
Put it in front of 20 real people. Past guests, your Instagram followers who've asked "do you do private dinners," and your catering list.

The payoff: quoting moves out of your texts at 11pm and into a page that quotes itself.

For consultants

Every new client starts with the same free 45-minute call, and I still don't have a page that would let them just book and pay for it.

The flagship diagnostic page

Tool: Claude Artifacts or Gamma for the page · Stripe Payment Links for the diagnostic fee

Pick the thing. Turn the free call into a paid diagnostic session, priced once, with a real deliverable attached.
Gather the raw copy: what the diagnostic covers, what they leave with, and two case studies (client names optional).
Have AI build the page. Describe the offer to Claude Artifacts and get a page that reads like a real service, not a lead magnet.
Add the payment link. A Stripe Payment Link for the diagnostic fee, due at booking.
Put it in front of 20 real people. Your warmest pipeline contacts and the last clients who said "I wish we'd started here."

The payoff: the free call that ate your calendar becomes the first paid deliverable, and it filters for people ready to buy.

For health advocates

Families ask the same five questions before they ever become a client, and I answer them one exhausted phone call at a time.

The resource guide page, privacy first

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links for checkout · no client details in the builder, ever

Draw the line first. No client or patient information goes into a general AI tool. This page is built from your own general process, never a case.
Gather the raw copy: the five questions you always get, your services, and your fee structure in plain language.
Have AI build the page. Describe the guide to Gamma and get a page that answers those five questions before the first call.
Add the payment link. A Stripe Payment Link if the guide or a paid navigation session has a price attached.
Put it in front of 20 real people. Referring providers, your waitlist, and the support groups where families already find you.

The payoff: the first exhausted phone call gets replaced by a page that already did the explaining, with the privacy line intact.

For makers

My next collection is finished before I ever get around to building the page that would actually sell it.

The drop page

Tool: Gamma for the page · Gumroad for the drop

Pick the thing. One drop, a fixed number of pieces, one page, one date, instead of an always-open shop nobody visits.
Gather the raw copy: each piece's price, materials, and the two sentences you'd say about it in person.
Have AI build the page. Feed Gamma your photos and details and get a drop page back with a countdown to the release.
Add the payment link. Gumroad checkout per piece, live the moment the drop opens.
Put it in front of 20 real people. Your waitlist, your best repeat customers, and the people who liked the sneak peek post.

The payoff: the collection that was already finished finally has a release day, not just a shelf.

For nonprofit leaders

Our biggest campaign of the year still runs through a link I made in 2019, and it shows.

The campaign page

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links or your donor platform's checkout

Pick the thing. One campaign, one goal number, one deadline, instead of a general "donate" button with no story attached.
Gather the raw copy: the goal, the impact per dollar if you have it, and two real program details.
Have AI build the page. Describe the campaign to Gamma and get a page with a clear ask and a progress bar toward the goal.
Add the payment link. A Stripe Payment Link or your existing donor platform's checkout, whichever your board already trusts.
Put it in front of 20 real people. Your board, your top past donors, and the volunteers who share everything you post.

The payoff: the campaign gets a page that matches the mission, not a leftover link from six years ago.

For photographers

Every mini-session season I rebuild the same booking spreadsheet instead of the page that would run itself.

The mini-session booking page

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links for the session fee

Pick the thing. One mini-session package, a fixed time slot length, price, and number of available slots.
Gather the raw copy: the price, what's included, turnaround time, and three sample images.
Have AI build the page. Describe the session to Gamma and get a page with the gallery, the price, and the slots laid out.
Add the payment link. A Stripe Payment Link that reserves the slot on payment, no back-and-forth needed.
Put it in front of 20 real people. Past clients from last season, your email list, and your last mini-session gallery's comments.

The payoff: the spreadsheet retires, and the page fills the slots while you're behind the camera.

For realtors

I know this neighborhood cold, and I've never once turned that into a page that brings people to me first.

The neighborhood guide page

Tool: Gamma or Claude Artifacts for the guide · your CRM's lead form for the opt-in

Pick the thing. One neighborhood, the one you actually farm, not a generic citywide guide.
Gather the raw copy: schools, transit, three favorite local spots, and honest current price ranges.
Have AI build the page. Describe the guide to Gamma and get a page that reads like a neighbor wrote it, not a brochure.
Add the link. A lead-capture form tied to your CRM in exchange for the full guide download.
Put it in front of 20 real people. Your past clients in that area, your sphere, and your last few open house sign-ins.

The payoff: the expertise you already had becomes a lead magnet that works while you're at a showing.

For stylists

New clients still find me through a DM back-and-forth instead of a page that would just let them book.

The book-me page

Tool: Gamma or Claude Artifacts for the page · Stripe Payment Links or your booking software

Pick the thing. One signature service, priced clearly, as the front door instead of a full menu that overwhelms a new client.
Gather the raw copy: the price, what's included, how long it takes, and three photos you're proud of.
Have AI build the page. Describe the service to Gamma and get a page with the price and the booking flow spelled out plainly.
Add the link. Your booking software's link, or a Stripe Payment Link for a deposit that holds the appointment.
Put it in front of 20 real people. Your current clients (they refer), your Instagram bio link, and your last month of DM inquiries.

The payoff: booking moves out of the DMs and into a page that works at 1am, when new clients are actually scrolling.

The house rule

Lauren's line still holds here: what belongs to you and keeps your humanity, you keep. The offer, the price, the way you talk to your buyers, that's yours, always. What frees you to be more human away from your laptop, you hand off, and the page that holds your offer is exactly that kind of hand-off now. If you started here and skipped Lesson 01, the whole loop makes a lot more sense once you've met the agent underneath it.

Find your lane at the party →