KPacks

At a glance

• An assembly line that turns an expert’s newsletter, book, or course into checklists, rubrics, and playbooks you can actually use.
• Born from a reading pile I never revisited. The question shifted from “how do I remember this?” to “how do I turn it into a tool?”
• Around fifteen packs so far, including ones built from Lenny Rachitsky’s data and Matt Mochary’s curriculum.
• Just markdown files and a couple of Python scripts. Any assistant that can read a directory runs the whole thing.

The reading pile

I’ll admit it: for years I collected things I never used.

Newsletters. PDFs. The odd book someone swore would rewire how I thought. I’d read something sharp on a Tuesday, nod along, feel a little smarter, and by Friday it was gone. Highlighting isn’t a system. Notes aren’t either. I’ve got folders of them I have never once opened again. (You probably do too.)

So sometime around August of 2025 — in a conversation with ChatGPT, of all places — I started poking at a different question. What if I could feed a focused pile of related knowledge into something and get a genuinely more useful form out the other side? Not another summary. A tool.

I took the idea to Claude, dumped in a stack of marketing and AI-marketing documents, and called the result my Growth Intelligence Studio. What came back surprised me: a rich, cross-linked set of derivative assets that behaved like a hyper-focused, very recent slice of training. And I could sit down and chat with it.

That was the moment the question flipped. It stopped being “how do I remember this?” and became “how do I turn this into a tool?” Those are very different questions. The second one has an answer.

What a kpack actually is

Drop in something an expert wrote — a newsletter issue, a book chapter, a course transcript — and out comes a set of things you can use on a real business. Checklists. Scoring rubrics. Worksheets. A short playbook. A prompt that turns the whole pack into an agent.

The metaphor in the repo is mining, and it earns its keep. Raw rock goes in. Metal comes out. Nobody builds anything out of rock.

Under the hood it screens every source first (most content doesn’t clear the bar, sorry), then mines out the frameworks and mechanisms, sorts them, and traces the cause, the effect, and the one place you can actually push. One rule runs through all of it: ideas, never wording. You can learn how somebody thinks. You can’t take their sentences.

Where it’s gone since

I turned the whole thing into a template repo, so I can go from zero to a useful kpack in an afternoon. I’m at maybe fifteen of them now.

When Lenny Rachitsky opened up LennysData.com, I pulled the articles and built a pack from them. (He’s since hinted at an MCP chasing something in the same spirit. Good — it’s a real problem.) I did the same with selections from Matt Mochary’s curriculum. So now, when a question’s rattling around my head, I fire up Claude inside one of these repos and just talk to it. Plenty of the answer comes from the model’s training. But plenty comes from the kpack — its context, its structure. That’s the whole point.

The most expensive lesson so far? Cluster before you process. Ten articles about the same mechanic aren’t ten tools. They’re one good one. A batch that looks like 300 articles is usually 30 or 40 real modules once you group them, and figuring that out before you spend a single token is the highest-leverage move in the pipeline.

There’s no app under any of this. It’s markdown files and a couple of Python scripts, and any assistant that can read a directory can run the whole thing. It’s just a repo. Which, honestly, is exactly how I wanted it…