We’re All Just Making This Up As We Go Along

And that’s not just a good thing, that’s the point

I kneecapped my Claude Writing Project about thirty minutes after I gave it an “upgrade.”

Here’s what happened.

After Part 2, I was enjoying having a writing system that actually seemed to work for me—Skills built based on Nick Quick’s VAST framework (Vocabulary, Architecture, Stance, Tempo), a Millie persona that gave real feedback, outputs that sounded like me.

Good stuff.

Skills and Prompts, oh I get it now

Skills And Prompts: Part Two—Millie Gets Down to Work

But as I corrected and adjusted the output I wondered: “Would it know for the next time?” Yes, I know Claude has Memory, but I wanted to make sure specific work, in a specific project, really stuck.

I’ve been watching and testing Jeremy Wright – Marketer/ECHO his system reads and writes core files from a private GitHub repository. That’s how it learns. Not context window memory. Not vibes. An actual place to put things down and pick them back up.

So I built the same thing into my writing project. Claude takes notes as we work. Periodically we review them and update my Voiceprint.

Iterative. Smart. Adaptive.

Except I’d just broken the whole thing without realizing it.

GitHub MCP servers don’t work on mobile—which meant my Voiceprint lived somewhere Claude couldn’t reach. No context. No voice. Pick up my phone, open the writing project, and Claude had no idea who I was or how I wrote. I’d just turned a writing partner who I could chat with using voice mode on my phone into a brick.

Twelve hours later, the fix hit me with that blinding flash of the obvious.

Oh. Right. Project files.

I can just put the Voiceprint and session notes directly in the Claude Project and sync them manually when I’m back on desktop.

Duh. ?????

I’m writing this on my phone right now. After making two small tweaks to the instructions. Which is either poetic or embarrassing—honestly, maybe both.

Here’s the thing about all of this: nobody has figured this out yet.

Not really.

Kim Doyal built a personal AI project system that works brilliantly—for Kim. Jeremy built ECHO for Jeremy (I like it too, btw). I built something that immediately broke and needed a patch.

I Stopped Using Other People’s Systems. Here’s What I Built Instead [VIDEO]

That’s not failure. That’s progress.

We’re so early in this whole AI thing (waves arms in the air), that the scientific method is the only way here.

F* around. Find out. Iterate.

The people who look like they have it figured out are just a few iterations ahead—and half of them hit the same dumb walls you will.

Or they are f*cking liars/shysters.

The spark isn’t making the perfect system. The spark is realizing you can build a system in the first place.

Then make it better.

Then fix the thing you broke.

Part 3 is coming—and it picks up right where Millie left off. Skills? Projects? Prompts? (Oh my!) When do you use which? I’m still figuring that out too, in real time apparently, but first I needed to tell you this part—the part where I built something, broke it, fixed it, and wrote the post about it while the fix was still warm.

That’s how this works now. Now go forth and let the f*ckery and finding out commence!