If last post was the thesis, this one is the receipts.
Here’s what actually happened this week across my active projects, and the specific AI workflows I used to get it done.
GoFundMe: Donor Pattern Analysis and Outreach
My campaign has been running for a few weeks now, and I’ve been using Claude to help me do something I couldn’t easily do manually: look for patterns in who’s donating, when, and in response to what.
The workflow is simple but effective. I export donor data, drop it into a conversation with Claude, and ask it to surface patterns — timing relative to posts I made, average gift size, whether first-time donors tended to come from a specific channel. Then I use those patterns to adjust my outreach.
What surprised me this week: Claude flagged that my update posts that included a specific personal detail (not just stats, but a real moment from what I’m going through) were generating significantly more shares than the ones that leaned heavily on the numbers. I knew this intellectually — that emotional resonance drives sharing — but having it reflected back as a pattern from my own data made it feel different. More actionable.
I’ve also been using AI to draft donor acknowledgment messages that feel personal rather than templated. I feed in the donor’s name, the amount, any context I have about them (did they leave a note? Do I know them personally?), and ask Claude to draft something warm and specific. Then I edit. The editing takes maybe 30 seconds per message instead of three minutes. That’s not trivial when you’re doing it at volume.
transkrybe.com: A Bug I Almost Shipped
I’m building transkrybe.com as a MusicXML transcription tool — it takes audio input and produces professional-grade sheet music. This week I was working through a bug in how the output was handling polyphonic passages (multiple simultaneous notes), and the way I debugged it is a good example of what I mean when I say I’m running this from my phone.
I described the bug to Claude in plain language: “When there are two notes played at the same time, one of them is sometimes dropping to the wrong octave in the MusicXML output.” Claude walked me through three possible causes, ranked by likelihood. The second one was right. I fixed it without ever opening a desktop IDE.
This is the thing I want executives to understand: you don’t need to be a developer to work with developers more effectively — and increasingly, you don’t need a developer for a lot of what used to require one. I’m not saying AI replaces engineers. I’m saying the threshold for what a technically-curious non-coder can ship has moved dramatically.
Job Search: Building the Pipeline
I’m not applying for jobs the traditional way. I’ve built a lightweight system that does most of the scouting for me.
The core of it: I’ve defined a target role profile (AI-forward CMO, CRO, VP of Growth, or fractional executive roles at companies actively building AI products). I use Claude to evaluate job postings against that profile — not just the title, but the actual language of the description. Does the company seem to understand what AI-forward leadership means, or are they just adding “AI” to a traditional job description? That distinction matters a lot to me.
When a posting passes that filter, I draft a tailored cover letter starting from a template I’ve built and refined over a dozen iterations with Claude. The key insight I’ve had: the cover letter should do one thing — demonstrate that I understand their actual problem, not just their job description. Claude helps me find that angle quickly.
I haven’t landed anything yet. I’m being selective. But the pipeline is organized in a way that would have taken me twice as long to build manually — and I can update it from anywhere.
What I’d Do Differently
One thing I’ve learned this week: the agents are only as good as the context you give them. I had a few conversations go sideways because I gave Claude too little background on a situation and it made reasonable assumptions that weren’t right for my specific case. The fix is simple — more upfront context — but it’s a discipline that takes practice. I’m still building the habit.
Next Week
I’m going to dig into the Slack agent pipeline in detail — how I built it, what it monitors, and how I’ve structured it to actually reduce noise rather than add to it. Plus: the first real revenue numbers from transkrybe.com.
—
Jesse Stay is CEO of Stay N’ Alive Productions and the author of 13 books on social media and technology. Follow this series at staynalive.com/blog.
