Everyone’s out here posting about AI-generated art exhibitions and autonomous agents that can negotiate your salary. Meanwhile, I’m over here using it to check my kid’s math homework.
And honestly? I think I’m winning.
Here’s my boring AI week:
The 100-Slide Operating Review Deck
A few months back I was obsessed with the idea of generating usable powerpoint decks from data. Could I take a CSV or JSON file and turn it into a well styled PPTX without having to painfully move text boxes and change font sizes? Then Claude released their document skills and it seemed so much easier!
Now I have an agent that does it in less than 10 seconds! Hot take: maybe we’ve been measuring AI productivity wrong. Forget “lines of code generated.” Give me “slides per hour.” That’s the metric that matters to anyone who’s ever stared at a blank PowerPoint at 9 PM on a Sunday.
Vibe Coding My Way Through 6th Grade Math
Admittedly I was suffering from cognitive overload when my 12-year old asked me to review her homework. 36 algebra proofs that needed to be evaluated. I did the first 6 and tapped out! Rather than admit defeat to a 12-year-old, I “vibe coded” my way through it—which is a fancy way of saying I used Gemini and NanoBananaPro to solve the proofs and generate a color-coded answer key that let me walk through the solutions while pretending I was just “double-checking my work.”
Was this cheating? Unclear. Did my child learn something? Also unclear. Did I maintain my fragile mental state for one more day? Absolutely.
The Azure DevOps Backlog Audit (Without Ever Opening a Browser)
While I appreciate the value of SDLC tools, Azure Dev Ops is not my favorite UX to say the least. What are the trends on velocity? Are we closing more bugs than we’re opening? Can I see the distribution of completed features by business function, by quarter? All very doable in ADO, though equally annoying. Solution? Create a subagent that specializes in the ADO API and allow the user (me) to prompt it using natural language to generate such output right from the terminal.
The PRD That Wrote Itself
Okay, this one’s actually cool, and I’m not going to pretend otherwise.
I had a Teams meeting with one of our ops leads, turned on Facilitator, and fed the transcript to Claude and asked for a PRD. What came out wasn’t just a summary. It was a structured product requirements document with user stories, acceptance criteria, identified stakeholders and—here’s the kicker—a section on “Open Questions” that captured exactly the things we’d talked around but never resolved. It took three minutes. The meeting took sixty. I’m not saying AI should replace meetings, but I am saying AI should definitely replace the two hours of document-writing that usually follows them.
The Point (If There Is One)
Here’s what I’ve realized: the boring stuff is the stuff that matters. If we’re talking about places AI can make an impact, it may also be the most accessible target for most people.
Nobody’s going to write a breathless LinkedIn post about using AI to check a backlog or help with homework. There’s no TED talk in “I didn’t have to make a PowerPoint from scratch.”
But that’s where the hours actually go. That’s the death-by-a-thousand-cuts of knowledge work—the slide formatting, the ticket grooming, the translating of meeting chaos into actionable documentation.
AI didn’t make my week exciting. It made my week efficient. I got those hours back. I used them for actual thinking! (And also for writing this blog post, which is arguably not thinking, but here we are.)
Next week? Expense reports!