I hit a wall I didn’t see coming.
It wasn’t burnout or writer’s block. It was stranger. I’d built a full year of technical playbooks, prompt systems, and vibe coding workflows right here on Substack. And I couldn’t find any of it.
Every time I started a new build, I knew I’d already solved the problem somewhere in my archive. So I’d spend 20 minutes scrolling. Then another 10 in Craft.do, where I had 180+ fragmented notes that were somehow even less organized than the newsletter. Then I’d give up and reinvent the wheel.
That’s when I realized the newsletter format had become a trap. A linear feed is great for discovery. It’s terrible for operational use. Every post disappears into the scroll the moment the next one drops.
So I stopped shipping any new feature for 72 hours and started dismantling.
The Great Substack Dismantling
A few weeks ago, I decided to stop searching and start architecting. I realized that if I wanted to stay true to the “Prompt-Led Product” standard, I had to treat my own work as an operational engine, not a blog.
I spent the last few weeks “in the studs,” tearing my own archive down to the foundation. Substack offers an export of all your writing, so I took that as the starting point. In my mind, all the valuable lessons were there, right? So I immediately thought:
I have all the data already.
I can run a Python script to condense everything into a new CSV.
This new CSV is my database for everything. Now I can tell AI to analyze all my content and generate valuable pieces from it.
The data migration was a disaster. I tried to export my work from Substack, thinking I could just “port” it over. I was met with a graveyard of 180+ massive, messy .html files. The formatting was broken, the code was mangled, and the logic was buried under layers of newsletter fluff.
💡 Pro Tip: LLMs have a “context window,” but they also have a “concentration window.” When you shove massive amounts of data through a script, the model starts to normalize everything. It loses the nuance. It sees a 2,000-word post and tries to summarize it into a 50-word CSV cell.
A script can move text, but it can’t perform a logic audit. I realized this wasn’t a data migration problem, but rather a curation problem. I couldn’t just copy-paste my way to a solution. I had to dismantle a year of my writer life.
I manually audited every single one of those 180 workflows. I stripped away the “Substack styling” and extracted the raw architectural components that actually matter: The Problem, The Logic, The Prompt, and The Strategy. It was tedious, frustrating work. But it was the only way to move from a “Content Archive” to an “Operational Engine.”
I Break AI Tools So You Don’t Have To
While I was tearing down my own system, I realized I should be doing this for yours, too. I’ve officially started doing Product Audits.
If you’re building a product and want a Senior PM to find the exact friction point where your users give up, I’m opening up slots. I’ll audit your prompt architecture and stress-test your logic before your users do. You can book a Product Audit here.
Welcome To The New Prompt Led Product Premium Vault
If you are a free subscriber, you’ve seen the theory. But if you want to see the exact systems, logic, and frameworks I use to fix the exact moment users quit, and the exact prompts I used to code this new platform, it’s time to step inside the engine.