AI writing's moat isn't the model, it's the proprietary data you feed it
Every writer Katie spent two years turning ChatGPT from a career coach during a layoff into a one-person content company and a compounding editor; the core method isn't prompts, it's using proprietary experience to cover the "last mile" the model can't reach.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The biggest value of an AI coach isn't saving money
After being laid off she was at a low point and couldn't afford a human career coach's hourly rate, but a $20-a-month ChatGPT subscription was acceptable. She found that writing her thoughts out and letting AI question and push back from the outside helped her externalise her thinking, counter catastrophising, and gave her a sense of accountability. The most important legacy of that experience wasn't the money saved, but that it made her believe AI really can drive concrete change in a life. At the time she was hesitating over whether to take Every's freelance offer, and it was ChatGPT that gave her the push to just try it.
— Katie ParrottThroughput comes from upfront context, not clever prompts
She once took on a delivery load of 8 blog posts, 3 ebooks, 24 LinkedIn posts, 24 X posts and 16 Instagram posts in a week and a half. She could do it not because she knew how to write a smarter one-line prompt, but because she did a great deal of upfront context engineering: feeding the model brand information, product details, audience profiles and differentiators all at once. Much of the demand was essentially content repurposing, and AI is good at rewriting one long piece into versions for different platforms. The rule she distilled: set the scene up front, and only then can you move fast later.
— Katie ParrottGive the model a fence first, then talk tone and word choice
Before Claude Projects existed, she could only keep persistent documents in Google Docs and manually copy-paste them into context each time. This structure didn't fall from the sky; it comes from the style guide tradition in content marketing: first define who the audience is, what the pain points are, how the product maps to them, who the competitors are and where the differentiation lies. She says this amounts to giving the model a playground with a fence: first make sure it won't wander off, then talk about word choice, tone and reading difficulty. Later, as soon as Projects launched, she built a separate project for every client and every column.
— Katie ParrottAI writing's last mile can only be covered by a human
She believes there is a "last mile" problem in AI writing: the model has a knowledge cutoff, and it isn't in the real physical world. So the human's job is to supply the latest data and real experience that AI can't reach. She warns that if you just throw the model a single line like "write a blog post about style guides", all you'll get is generic content it already knows, which is commoditised information; what really makes writing distinctive is fresh ingredients the model has never seen, such as internal company research, third-party reports and personal experience. The writing system is the kitchen; the human is responsible for providing good ingredients.
— Katie ParrottAI's value is reducing friction, not speeding up output
She talks openly about her bipolar disorder and points out that AI's value isn't just producing content faster, but reducing friction in life. She put off booking a primary care doctor for three years, and in the end just had Codex find doctors who accepted her insurance and were taking new patients and complete the booking; email is first filtered by automation, and only messages that genuinely need a human reply reach her inbox. She says this completely solved her email anxiety. This isn't treating AI as a production tool, but as assistive technology that makes her day-to-day running "as a person" easier.
— Katie ParrottLet AI interview you repeatedly, and it can prioritise for you
Two years ago she was just asking ChatGPT whether it could be her career coach; today that coach lives in a Codex project containing her role profile, Every's brand positioning, content performance data, a folder of reader praise and her OKRs. She doesn't maintain a board herself; she uses voice conversations to have the system prioritise by highest impact, and the system also maintains her Kanban for her. The key to building this system was letting AI interview her repeatedly and then sedimenting the interview content into context. She says it's like having a Chief of Staff, not just a writer.
— Katie ParrottFeedback that only fixes the current draft is wasted
The core of the Compound Writing plugin is compounding thinking: every piece of feedback given to AI should flow back into the system so the next output is automatically better. She forked Kieran Klassen's Compound Engineering plugin and reworked it into a writing workflow: code and writing both have brainstorming, planning, drafting and review stages, except writing focuses more on structure, voice and evidence. Editing is further split into a substantive edit that looks at big structure and argument, a line edit that works sentence by sentence, and a final pass that checks before publication. Now she doesn't write in a web chat window, but feeds the day's fresh material into this system.
— Katie ParrottVonnegut as editor gives you a perspective, not a verdict
She built Vonnegut's eight narrative elements, Hitchcock's principle of suspense about the bomb under the table, and the expressive styles of Sorkin and Sedaris into the plugin as editing skills. Vonnegut checks whether "the story starts as close to the end as possible" and whether "every sentence earns its place"; Hitchcock's perspective asks whether the reader will want to keep reading because of this. She stresses that this isn't letting the masters' ghosts grade your draft, but giving you a new perspective you can accept or reject; non-professional writers may actually benefit more, because they're less likely to be picking at themselves while they write.
— Katie ParrottAI's compounding may only compound for the few who got in early
She says her core thesis about AI today is this: access to education and to using it will matter more than ever. She lives in Columbus, Ohio, and has never lived in New York or San Francisco, which makes her care more about how opportunity is distributed. She admits she got where she is because she had financial slack, a time budget and a network with access to excellent peers; she hopes the AI community will actively create conditions for more places and more kinds of people to get on board. Because she sees clearly that AI has enormous compounding effects, and the cost is that the returns may only compound for the few who happened to get in early.
— Katie ParrottIn their own words · checked verbatim
Clarity, it turns out, doesn't arrive gift wrapped from a digital assistant or even a human coach. It's something I had to dig out for myself, question by question and prompt by prompt.
Natalia Quintero5:13
And our job as the humans is to close that last mile and provide the real world experiences that AI can't get because they come after the knowledge cut off and it's also out there happening somewhere in the real world and AI is not in the real world, the physical world yet.
Katie Parrott15:40
We think about AI as a productive technology, right? We value it for its ability to produce output and get things out of our, like just put stuff out. And what I found in my personal experience is that AI is just as powerful as a supportive technology.
Katie Parrott22:50
compounding is the idea that every piece of feedback that you give to AI should feed back into the system to improve the next output.
Katie Parrott32:27
My thesis on AI today is that education and access are going to matter more than they ever have before.
Katie Parrott44:49
But the risk there is that value is going to compound to the small subset of people that happen to be early. And I just think that a much more compelling vision for an AI future is one where everyone is able to come along for the ride.
Katie Parrott46:32
Figures
| Human career coach hourly rate vs ChatGPT monthly fee | about $150/hour vs $20/month | 3:11 |
| Time spent putting off booking a primary care doctor | about 3 years | 23:53 |
| Number of security vulnerabilities Codex found in her Tastemaker project | 5 | 43:24 |
Glossary
- context engineering
- Pre-organising information such as brand, audience, product and differentiation into a fixed background so every AI conversation heads in the right direction.
- last mile problem
- Model training has a knowledge cutoff, so humans need to supply the latest data and real-world experience for writing not to fall into generic platitudes.
- compound engineering
- Sedimenting one-off feedback given to AI into the system so the same good decisions automatically take effect again and again in future.
- computer errands
- Having an AI agent handle administrative chores you keep putting off, such as booking doctors, filing insurance claims and sorting your inbox.
How to listen
Independent writers and content entrepreneurs struggling with the quality of AI-generated writing, and anyone who wants an AI agent to take over life and administrative tasks.
The layoff story and the ad break in the first few minutes can be fast-forwarded; start on the methodology at the 8:26 delivery case.