LOADING THE FEED ▮
NICHE OF ONE
--:--
← The Feed

How to Use AI Tools Without Betting Your Creative Business on One Company

/Skip the AI panic and the AI hype. How independent creators can use AI tools for grunt work, avoid platform lock-in, and keep a way out when prices change.

post to X email it
Flat black silhouette of a hand pulling a lever on cream with a halftone dot field, the Niche of One card for this post.
// the field guide Empire of One Build a creator business at human scale — niche, list, money, kept straight. $7 →

TL;DR: The AI fight has three camps, and none of them is working for you. Use the tools for grunt work, keep the real work yours, and build so no single company can switch you off.

Everybody yelling about AI right now is selling something. The big companies sell licensing deals and lobbying. The hype guys sell courses. The angry crowd sells the feeling of being on the right side. I don’t belong to any of them. If you’ve read me for a while, I’d bet you don’t either.

So here’s what’s left when you tune all three out.

Why isn’t protesting AI a working strategy for creators?

The money already moved, and it went to whoever had lawyers on retainer. I’m no tech utopian. I’ve watched writers I respected put twenty years into their craft, then watched a language model cough up something half-decent in seconds. I got the same sick feeling they did. That grief is real.

Now follow the money:

  • Publishers signed licensing deals. News Corp, Axel Springer, the Associated Press, Condé Nast and others licensed their archives to AI companies.
  • Platforms sold what their users wrote. Reddit users wrote the posts. Reddit got paid by Google and OpenAI. The users got the terms of service.
  • The biggest author settlement was a price tag. Anthropic agreed to pay $1.5 billion over nearly half a million pirated books, and a federal judge gave it final approval in July 2026. That’s roughly $3,000 a work before lawyers’ fees and publisher splits. The court had already ruled that training on books Anthropic bought legally was fair use. The piracy cost money. The training stayed legal.

Rules that only the biggest companies can afford to follow end up protecting the companies that already signed. The guy at the kitchen table gets nothing out of them. Refusing to touch AI is a fair choice, and plenty of good people make it. It still isn’t a plan for paying rent.

Why isn’t the AI hype pitch true either?

The tool makes you more of what you already are. If you’ve got skill, taste and something to say, it gets the grunt work off your desk. If you don’t, you crank out the same mediocre stuff the internet is already drowning in, only faster.

Nobody can sell that as a $29 prompt pack, so they sell you a magic sentence instead. There isn’t one. Any prompt a course charges for, you can figure out yourself in an afternoon of messing around.

The tools are also wildly uneven. A model will knock out a hard problem and then faceplant on something a kid would get right, and you can’t tell in advance which one you’ll get. You find out by running it and checking the output. A human editor stays in the loop for that reason. Anybody telling you to fire the editor and let the machine drive is selling you a fantasy.

What do platforms actually do once you depend on them?

They raise the price and change the rules, usually in that order. It’s the oldest move in the creator economy, and AI companies run it too.

Here’s where some of the platforms sit as of November 2026:

  • Gumroad: 10% plus 50 cents on direct sales, with card processing on top. Sales that come through its Discover marketplace cost 30%.
  • Patreon: a flat 10% for anyone who published their page after August 4, 2025. Unpublish an older page, even briefly, and you lose your old rate.
  • Substack: 10% of paid subscriptions plus Stripe fees. Your email list comes with you if you leave. Moving paying subscribers runs through Stripe, so learn how that works before the day you need it.
  • Facebook: in 2018 it reworked the feed to favor friends and family. Publishers that had rebuilt their newsrooms around Facebook video watched the traffic fall off a cliff. Slate reportedly lost 81% of its Facebook traffic in a year.

AI vendors follow the same staircase. Free tier, paid tier, pricier tier, then “contact sales.” The API gets metered and the limits move without much warning. Building your whole business on one vendor’s current pricing is the same bet those publishers made on Facebook.

Rented land is fine for reach. Use it. But write your exit plan before you need it, and keep your own site, your own list and your own copies of everything.

Can you run AI tools without paying a big vendor?

Yes, and it’s a lot easier than it was two years ago. Open-weight models you download and run on your own machine handle plenty of narrow jobs now: drafting, summarizing, formatting, code review, pulling facts out of documents.

  • Qwen is Alibaba’s model family. The open releases through Qwen 3.6 ship under Apache 2.0, which allows commercial use. Qwen 3.7 launched API-only with no downloadable weights, so read the license on the exact model before you build anything on it.
  • Ollama and LM Studio are the easy on-ramps for running these on a regular computer. Pick a model that fits your hardware and find out what it can and can’t do.
  • Cheap hosted APIs for open models can cost a fraction of what the big labs charge. One consulting firm said it moved a job classifying 50,000 documents a day from GPT-5 to DeepSeek V4 and the monthly bill dropped from $4,200 to $210, with accuracy within two points. That’s one firm’s word. Test your own workload before you trust the math.

I’m no China fanboy, and I doubt you are. Some of these labs have murky training data and politics worth a hard look. If you send your queries to a hosted API, find out where it runs. If the weights sit on your own hardware, that question mostly goes away. Either way, a second vendor won’t make any tool safer, but it does make you a lot harder to hold hostage.

What should an independent creator actually do?

Use the tools for narrow jobs you already understand, and set things up so you can switch. In practice:

  • Keep the thinking yours. Let AI help with drafts, research, formatting and review. Keep the part that makes the work yours.
  • Check everything. The output is a first pass, every time.
  • Don’t wire your business to one provider. If switching vendors takes a weekend instead of a rebuild, you’ve got a say in the price.
  • Learn to run one local model. You won’t need it every day. You’ll need it the day the pricing page changes.
  • Own the audience. Your email list, your domain, your archive.
  • Treat AI creators as peers. Somebody using these tools tells you nothing about whether they have something to say. Plenty of the most interesting people building right now use AI every day, and they belong in the room.

There was a hardware store in the town I grew up in. You’d bring the old man a rusted fitting, he’d look at it for maybe four seconds, then walk you to an aisle you never knew was there. Forty years of plumbing lived in his head. The place closed during the Bush years. Lowe’s didn’t inherit a lick of what he knew.

That shop is what none of the three camps is fighting for. Somebody who knows one thing cold, and earns a decent living off the people who need exactly that, doesn’t figure into any of their plans. The tools can help that person. They just can’t be the foundation.

Frequently asked questions

Is it okay to use AI as an independent writer or artist?

Yes. Using AI for grunt work like formatting, summarizing and first-pass research is your call, and lots of working creators make it. What matters is that you still do the thinking, bring the taste and make the final edit. Tell your audience where they’d care to know.

Should I run AI models locally instead of paying for ChatGPT or Claude?

Run one alongside whatever you pay for. Local models cost nothing to run once you have the hardware, keep your data on your own machine, and can’t be repriced out from under you. They’re weaker than the best paid models on hard tasks. A sensible setup is a paid tool for the hard work and a local model for routine jobs, so a terms change doesn’t strand you.

Are open-weight models really free for commercial use?

Some are. Models released under Apache 2.0, including Alibaba’s open Qwen releases through 3.6, allow commercial use. Others carry custom licenses with limits, and some newer releases, like Qwen 3.7, have no open weights at all. Check the license on the specific model version you plan to use.

What are the fees on Gumroad, Patreon and Substack in 2026?

Gumroad takes 10% plus 50 cents on direct sales and 30% on sales through its Discover marketplace. Patreon charges a flat 10% to creators who published their page after August 4, 2025. Substack takes 10% of paid subscriptions. Payment processing comes on top in most cases, so check each platform’s own pricing page before you plan around a number.

Did the Anthropic book settlement pay authors?

A federal judge gave final approval to the $1.5 billion settlement in July 2026. It covers nearly half a million pirated books, at roughly $3,000 per work before attorneys’ fees and publisher splits, and payments had not gone out at approval. Authors whose rights sit with a publisher may see a smaller share. The settlement paid for the pirated copies. It did not stop training on legally bought books, which the court had already called fair use.

How do I avoid platform lock-in as a creator?

Use every platform to find readers and move them somewhere you control. Collect emails on your own list, publish on a domain you own, export your archive on a schedule, and learn how to move your paying customers before you have to.


Keep your own list and a local model that works, and the next pricing change is just an annoyance.

// comments
Full search on OneSearch: the network, the ring, and the open web →esc closes · ↑↓ move · ↵ opens