LOADING THE FEED ▮
NICHE OF ONE
LIVE --:--
Join free
← The Feed
a bright red EXIT door built into a towering wall of grey filing cabinets, one drawer hanging open and spilling papers onto the floor
Feature · ownership · money · ai

The Exit Strategy

Every platform you're renting on is running the same extraction playbook coal companies and cotton merchants ran a century ago. What's different now is that, for the first time, you don't have to.

J.D. Forrest · · 22 min read
// the everything pass All-Access The whole catalog, the members vault, and the back room where the operators talk shop. $67/yr →
post to X email it

Substack has a calculator built right into its own website, and the number it shows you is already an undersell. Feed it ten dollars a month and four hundred subscribers and it spits out six hundred thirty-six dollars gone before you see a dime. That figure covers the platform’s ten percent and Stripe’s standard card-processing cut, but Stripe also charges platforms a separate 0.7 percent recurring-billing fee on top, one Substack doesn’t headline.

Run the full math and the effective rate lands closer to sixteen and a half percent, not ten. Push it to ten thousand subscribers and you’re gone fifteen thousand nine hundred dollars a month. Get to fifty thousand, the kind of list a writer spends a decade building, and it’s seventy-nine thousand five hundred dollars a month, every month, forever.

Call it what it is. Almost a million dollars a year, gone to a company that never wrote one word of it.

That’s a documented business strategy, not a glitch in an otherwise generous system, and every “alternative” people flee to once they see it is running the same model with better manners. I’ll get to that.

The trap isn’t the fee

Here’s how it works underneath the fee schedule. When you show up new, the platform hand-feeds you: a featured slot, an algorithmic push, some borrowed traffic while you’re fresh enough to be a good growth story for their own metrics, the way a farmer talks sweet to livestock the week before slaughter.

Sean Highkin found this out running The Rose Garden Report on Substack. Featured early, funneled traffic, growing fast.

Then the day he stopped being “new recruited talent,” the featuring stopped and the growth flatlined right along with it. He moved to Ghost, cut his bill from $4,968 a year to $2,052, and by his own account makes more money now than he ever did on the platform that discovered him.

He’s not alone, and the list keeps growing. Matt Brown grew Extra Points to seventy-one thousand subscribers on Beehiiv. Run those numbers through Substack’s math and he’d owe twenty-five thousand dollars a year; on Beehiiv he pays about three thousand.

Ryan Broderick left Substack for Beehiiv in January 2024 and says what he used to pay “was not enough to hire a full-time employee.” Now he has one. Luke O’Neil was paying ten thousand dollars a year before he moved Welcome to Hell World to Ghost.

Marisa Kabas says her subscriber count tripled after she left, and that it felt good to stop handing over ten percent. Substack has roughly sixty-three thousand newsletter creators total.

The ones leaving are still a small fraction. Most people don’t leave. Most people don’t know there’s a door.

The real trap was never the ten percent, or the sixteen and a half once Stripe gets its cut too. It’s the discovery layer. You feed it constantly, every week, forever, or you disappear from it.

Substack’s own cofounder, Hamish McKenzie, has said publicly that “no walled garden would let you export your mailing list, content, and even payment relationships at any moment,” and that’s true, you can. In a separate note, he addresses the harder question directly: why you can’t take your followers with you, the people who found you through Notes and never actually subscribed to anything.

Those belong to the algorithm. You built the audience. The platform owns the relationship, and it knows exactly what that’s worth.

The system was built to require constant feeding, and it does its job.

You will always give a platform more than it gives back to you. That isn’t bitterness. It’s the mechanics.

Pure extraction, running exactly as built, and once you can see the gears you can’t unsee them in any version of this business.

Every exit leads to a smaller cage

So people leave, and they should. Know where you’re headed first. Beehiiv doesn’t take a cut of subscription revenue, true, but you need its Scale plan to unlock monetization at all, and that starts around forty-three dollars a month, climbing well into the hundreds as your list grows, more again if you want the branding gone too.

Ghost doesn’t take a cut either, but its entry Starter tier, eighteen dollars a month as of a July 2025 price change, won’t let you charge for anything at all. You need Publisher, twenty-nine dollars a month, and that only covers you to a thousand members; past that you’re into the Business tier, a hundred ninety-nine dollars a month at ten thousand.

Kit takes three and a half percent plus thirty cents per transaction on top of a plan that went up thirty-five percent in September 2025. Patreon, as of August 2025, charges every new creator a flat ten percent, the same number Substack charges, replacing a tiered 5/8/12 system that mostly rewarded creators who’d already made it.

Press Gazette modeled a mid-size publisher, a hundred thousand subscribers, ten thousand paying, and found Substack costs at least fifty thousand pounds a year more than the alternatives. Fifty thousand pounds a year is not a rounding error either.

Every one of these platforms is the same arrangement wearing a friendlier fee schedule, and you’re still building someone else’s product with your own labor, betting the new landlord won’t enshittify the place the way the last one did, which he eventually will, because that is what happens once the money behind any platform needs a return on it.

There’s a word for that now. Cory Doctorow coined it in a 2022 blog post about Amazon that’s since been named word of the year by dictionaries on two continents.

Enshittification: a platform is good to its users until it locks them in, then good to business customers at the users’ expense until it locks them in too, then it claws back everything for its own shareholders, because by then nobody can afford to leave.

His own description of the operators running it, from the book he later wrote about it: “Your job is to create as much value on that platform as possible. Our job is to harvest all of that value, leaving behind the smaller quantum of utility that will keep the platform from imploding.”

You don’t have to take his word for it. Watch it happen across every platform that’s ever hosted a creator.

TikTok launched its Creator Fund in 2020 promising a billion dollars over three years and paid out two to three cents per thousand views, as little as a thirtieth of what creators say they made per thousand views on YouTube. Kevin Yatsushiro hit three million views on one video in 2021 and got paid twelve dollars and fifty cents.

Twitch’s Partner Plus program, launched in 2023, offered elite streamers a 70/30 split above the standard 50/50, but capped it at a hundred thousand dollars a year, reverting streamers to 50/50 past that, and only dropped the cap in January 2024 after what Forbes called major pushback.

Twitter killed its entire third-party developer ecosystem in February 2023 with about a week’s notice, ending apps like Tweetbot and Twitterrific that had run for over a decade, and replaced free API access with a hundred-dollar-a-month tier that later doubled.

Instagram’s organic reach fell from nineteen percent in 2021 toward roughly three and a half percent by 2025, and Instagram’s own head, Adam Mosseri, explained why the algorithmic feed exists at all: people were “missing 70 percent of all their posts,” so Instagram built a system to decide which thirty percent you’d see. Deciding what you see is deciding who gets seen, and that decision is now a service Instagram sells back to you.

Medium is maybe the cleanest case, because it has rebuilt how it pays writers again and again since 2012 and never landed anywhere for long. Ad revenue, then publisher partnerships that collapsed and blindsided the publishers who’d built businesses inside them, The Ringer and Backchannel among the casualties, then the “clap economy,” where one writer documented three dollars and one cent for five claps against eleven cents a clap elsewhere in the same month, because nobody outside Medium could see the formula, then reading-time-based pay, then a 2023 overhaul that made writers pay five dollars a month to earn anything at all. Publicly available data from before Medium stopped disclosing partner earnings in 2019 showed ninety-four percent of Partner Program writers earning less than a hundred dollars a month.

I know the next part personally. I went from a steady three to four hundred dollars a month on Medium to pennies overnight, and Medium announced its first profit in twelve years not long after. Thousands of other writers reported the same collapse in that same stretch.

Medium’s leadership calls that a coincidence. It goddamn wasn’t.

The math requires it

None of this needs a villain, and that’s the uncomfortable part. Doctorow’s enshittification names the shape. The venture capital behind most of these platforms explains why the shape is close to mandatory.

A typical VC fund needs a handful of enormous winners to cover a much larger pile of losses, what a16z’s own research calls the Babe Ruth effect: about six percent of investments, across the industry’s own data going back to 1985, generate roughly sixty percent of a fund’s returns.

Investors backing a newsletter platform aren’t funding a nice tool for writers. They’re funding a lottery ticket, and lottery tickets only pay out if the number gets big enough to matter.

Substack’s valuation jumped from six hundred fifty million dollars in 2021 to eleven hundred million by July 2025, a seventy percent leap, against reported annual revenue around forty-five million. Casey Newton, who covers this beat for a living, did the math publicly: venture investors “rarely invest in anything that they don’t believe has a significant possibility to return 10 times their investment.” Somebody has to close that gap eventually, and it isn’t going to be the fund eating the difference.

There’s economic theory underneath why the platform closes that gap on the creator’s side specifically instead of the reader’s. Jean Tirole and Jean-Charles Rochet worked out how a platform serving two different groups prices each side by how price-sensitive it is, not by fairness.

Their finding: the side that would leave if you charged it gets subsidized, and the side that’s stuck stays and pays. A reader can leave a free newsletter in one click. A creator with three years and thirty thousand subscribers built on one platform cannot leave nearly that cheap, so the platform prices accordingly.

It isn’t picking on you. It ran the numbers, and the numbers say you’re the side that stays.

Economists have a colder word for a market with effectively one buyer: monopsony, a term Joan Robinson brought into economics in 1933 to describe exactly the company-town arrangement two sections down. Suresh Naidu at Columbia measured it directly on Amazon’s Mechanical Turk, a supposedly frictionless online labor market with no geography at all, and found labor-supply elasticity around 0.1, meaning workers barely respond to being underpaid because leaving costs them almost everything they’ve built there.

Nobody has published that exact measurement for Substack or YouTube yet, and I’ll say that plainly instead of dressing up a hunch as a citation.

But the standard modern theory of monopsony holds that almost any market with real switching costs behaves this way, and an audience you spent three years building is exactly that kind of market. The theory generalizes even where the specific study hasn’t been run.

Older than the internet

None of this is a new shape either. Appalachian coal companies paid miners in scrip, company currency worth fifty to eighty cents on the dollar, spendable only at the company store, where prices ran higher than what the same goods cost in town.

Roughly three-quarters of all scrip circulating in America came from coal companies in Kentucky, Virginia, and West Virginia, and most miners never fully paid off what they owed the store before they died or moved on, which was rather the point. Merle Travis wrote “Sixteen Tons,” and its most famous line still describes exactly this, debt with no exit built into the system on purpose: “I owe my soul to the company store.”

George Pullman ran the identical logic through rent instead of scrip. Company housing in Pullman, Illinois was priced to guarantee the company a fixed return, and when the depression of 1893 hit, Pullman cut wages and laid off much of the workforce without lowering rent by a cent, deducted straight from what was left of the paycheck.

The company’s cut doesn’t move. Yours does.

Sharecropping ran the same machine with cotton. Tenant farmers had no cash after the Civil War, so merchants advanced food and seed against the coming harvest at interest rates economic historians Roger Ransom and Richard Sutch put as high as fifty to sixty percent annually, took the crop first when it came in, and left the farmer whatever remained, which was often nothing.

Ransom and Sutch called what resulted debt peonage, not hardship, because the institutions were built to produce exactly that outcome, and the trap caught white yeoman farmers alongside freedpeople in large numbers, not just the population the crop-lien system was first built to control. The lien didn’t check your race first.

By 1920, tenancy across the Texas blackland prairie counties ran as high as two out of three farmers, Black and white alike, working as tenants on land someone else owned, for a merchant who set the terms and held the lien.

A writer named Nicholas Carr saw the identical shape land on the internet back in 2006, years before Substack existed, and named it before anyone was calling it a business model: digital sharecropping. His words: “the sharecroppers operate happily in an attention economy while their overseers operate happily in a cash economy.”

He was writing about blogs and early social platforms, and the pattern he clocked then, a rising share of total traffic concentrating on a shrinking share of sites while the total number of sites kept doubling, is the same shape as fifty million people calling themselves creators while two million make a living at it.

The company changes. The interface changes. The machine that takes your labor and gives back just enough to keep you working stays remarkably consistent, whether it pays in scrip, in cotton-lien credit, or in the promise of going viral next week.

Nobody sat you down and explained the mechanism

The vocabulary changes by the time it reaches YouTube and TikTok, but the researchers studying it are describing the identical trap the miners and sharecroppers lived in, run through an app instead of a company store. Sophie Bishop at King’s College London calls the folk theories creators trade to guess what the algorithm wants “algorithmic gossip,” because the platform tells you nothing and you’re left swapping rumors with strangers about what might keep you visible.

Zoë Glatt’s research on YouTube creators found that they described “the algorithm” the same way, again and again: an omnipotent god, a black box to be opened, a mystery to be solved, a voracious machine, universally experienced as an antagonistic force.

Brooke Erin Duffy at Cornell has a name for the whole arrangement: aspirational labor, the unpaid or underpaid work millions of people do for platforms and brands on the promise that they might be the one who breaks through. Hope is the wage.

The business plan runs on volume, not on everyone breaking through, the same way a casino needs everyone at the tables so the one jackpot photo on the wall keeps the room full.

The numbers back the ethnography up. Fifty million people worldwide call themselves creators. Roughly two million do it full time.

That sliver accounts for most of the money in the whole economy built around this work. Everyone else is the traffic that makes the sliver possible.

You will always give a platform more than it gives back to you, and now there’s a body of research describing exactly how, with your name on it, more or less.

They’re slaves to the platforms is a hard thing to say about people you like. It’s also accurate, and it isn’t their fault.

Nobody sat these people down and explained the mechanism. The mechanism was built specifically so they wouldn’t notice it, one interface decision at a time, and it worked, because that was always the brief.

What’s actually new

What’s different now isn’t a new platform with a smaller cut. It’s that a single person, no funding, no engineering team, can build the entire apparatus themselves, and there’s finally hard data to say that’s not a fantasy.

METR, the group that actually measures this, tracks how long a task an AI model can complete unsupervised at even odds, and that number has doubled roughly every seven months since 2019, with the pace picking up further since 2024. On the industry’s own coding benchmark, models went from solving under two percent of real software bugs in 2023 to over seventy percent by mid-2025.

The tools doing that work, agents that plan a task, write code, run it, read the error, and fix it themselves, mostly didn’t exist as products before 2024: Devin arrived that March, GitHub’s Copilot Workspace that April, Replit’s Agent that September. Anthropic shipped a connector standard, Model Context Protocol, that November, so an agent can now talk to a payment processor or an email service without anyone hand-writing the integration code; Stripe built its own connector for exactly that. None of it existed in a form regular people could use three years ago.

Worth saying plainly, because I don’t want to oversell this either: the industry’s standard coding benchmark was caught with contamination problems in 2025, models reproducing memorized answers instead of solving fresh ones, so treat any single leaderboard number as a moving target, not gospel.

And wiring an AI agent straight into your payment processor carries serious risk if you don’t understand what you connected. Researchers have already documented ways to manipulate these agents through poisoned instructions hidden in ordinary-looking content.

This is powerful. It is not risk-free. Read the security writeups before you connect anything to real money.

A randomized study inside Google, ninety-six full-time engineers, measured AI assistance cutting time-on-task by roughly twenty-one percent, and that’s the number among people who already knew what they were doing before AI got involved. Extend that curve to someone starting from zero and the tool stops being optional. It’s the reason zero is a viable place to start from now.

Nieman Journalism Lab reported in December 2025 on what it called the rise of vibe-coding journalists, people with no development background using these exact tools to build what their newsrooms couldn’t afford to commission: “it’s now cheaper to build interactive digital things, and so many niche ideas that would otherwise be too expensive can now be built.” André Chaperon ran the actual math on his own migration off Substack: at two hundred sixty annual members paying two hundred fifty dollars each, sixty-five thousand dollars in revenue, Substack’s ten percent would have cost him sixty-five hundred dollars a year; his self-hosted stack, built with AI-assisted tools handling the parts he didn’t know how to code, ran him about eighteen hundred thirty-nine. Carrie Loranger runs a six-figure solo newsletter operation on a stack she says costs under a hundred seventy-five dollars a month total, using AI tools to build her own custom pieces instead of hiring anyone, replacing what she estimates would otherwise run seven to thirteen thousand dollars a month in salaries and contractors.

Here’s the part I won’t dress up: fully documented, ground-up, solo-built Substack replacements with verified independent revenue numbers are still rare. What is documented is solo operators using these exact tools to build serious paid infrastructure in adjacent categories. Dana Snyder built a subscription platform for nonprofits by herself over six months using AI coding tools, and she remains its only full-time employee.

That’s not a newsletter, but it proves the capability exists at the scale a solo creator would need. Most of what’s specifically newsletter-shaped out there right now is AI-assisted automation bolted onto an existing platform, not a wholesale replacement of it, and Fast Company is right to be skeptical of anyone selling vibe coding as a free pass past the hard parts.

Users you can’t just restart, money moving through a live payment flow, uptime nobody’s watching at three in the morning: those expose the gaps fast if you’re vibing your way through it without understanding what you built.

That’s worth admitting up front. I’d rather tell you now than have you find out at two in the morning when something breaks and you don’t know why.

But the cost math underneath all of it is solid, sourced, and boring in the best way.

A self-hosted Ghost install runs fifteen to fifty dollars a month once you account for the server and the email-sending service, with zero transaction fee on anything you sell through it, forever. A lighter setup, Listmonk for the mailing list and AWS SES for sending, plus Stripe wired directly to your own account for payment, runs eight to fifteen dollars a month from roughly a thousand subscribers up into the tens of thousands, because you’re paying for compute, not a percentage of your own success.

Set that against a platform fee that scales with your revenue, and the gap gets worse specifically the moment you start winning. Renting gets more expensive exactly when owning would have started paying for itself.

I run nicheof.one this way: a static site with payments and subscriptions wired straight to my own accounts, no platform in the middle taking a cut of anything. And I’m not guessing at what different configurations cost, because pricing out infrastructure at different scales is close to being part of what I do, so when I give you a range, I mean I’ve priced it, not glanced at a marketing page.

A bare-bones version of this, stripped to the studs, runs thirty to forty dollars a month. A more built-out version costs more, because I keep testing configurations most people never would.

But the floor holds, and it isn’t a secret. Five years ago that floor didn’t exist for anyone without a development team. Now it does.

The exception, and the objection

This is the one place I’ll tell people who can’t stand the idea of using AI anywhere near their creative work to make an exception. Use it here. This is what it’s for.

It writes the code you never learned. It debugs the part that breaks at two in the morning when nobody’s answering support tickets.

The fantasy of owning this instead of renting it used to cost forty thousand dollars and a computer science degree. Now it’s a weekend project, some documentation, and a willingness to sit with something until it works.

Somebody reading this is already building the objection. No money to build anything. Still need the discovery, still need the algorithm to hand them strangers while they’re new enough to be a good story.

Fine. Keep using it. That was always the honest use for these platforms in the first place.

Discovery, not shelter. Let them introduce you to people who’ve never heard of you. Stop paying rent to sleep there once the introductions stop mattering.

Fund the thing you own with the money the thing you’re renting generates. Thirty to fifty dollars a month is not a barrier at any revenue level worth having. It’s a rounding error against what you’re currently paying just to stay visible at all, and the gap between those two numbers is the whole argument.

The door

Some of you will read every word of this, every citation, every dollar figure, and keep paying anyway. Good. Nobody indoctrinated you into that.

You’ve got a genuine taste for simplicity you didn’t have to build with your own hands, and there’s nothing wrong with buying it instead. But the platform was never going to tell you the exit existed, because it was never in its interest to, and now you’ve read the receipts instead of taking my word for it.

Build the door before you need one. Whether you ever walk through it is yours to decide. Just don’t mistake the silence for there not being a door.

Sources

← More from The Feed
// comments
Open full search ↵esc closes · ↑↓ move · ↵ opens