Every note, in one place.
Some of these are standalone observations. A growing set of them belong to one thread — the AI Throughput research — and link to each other in sequence.
These started as pages like this one — before they became posts.
The Operational Bottleneck AI Still Can't Solve
After months of building, testing, and breaking systems, I've learned this: it's rarely the model. It's the handoff. The gap between automation and accountability.
Read Full Note →What Happened When I Let Two AI Agents Compete
Day 2 observations from the Battle of the Bots experiment — unexpected behaviors, emergent strategies, and what they reveal about autonomous decision systems.
Read Full Note →Why Most AI Demos Fail in Real Business Environments
The gap between sandbox performance and operational deployment is wider than most people admit. Here's what actually breaks.
Read Full Note →Building Trust in AI Systems One Decision at a Time
Accountability isn't a feature you bolt on later. It has to be designed from the start — visible, intentional, and human-centered.
Read Full Note →Why Every Improvement Creates Another Bottleneck
This is the note that started the rest of them. I kept closing out automation projects that worked exactly as designed, and the organization didn't get faster in any way I could point to.
Read Full Note →AI Is My Jonah
I said it as a joke in a chat window before I meant it. Somewhere in a few hundred back-and-forth exchanges, I noticed AI doing the same thing to me that Jonah did to Alex Rogo.
Read Full Note →Digital Work-in-Process
If AI can generate ten times the output, where does the tenth-times pile up? Manufacturing has a name for unsold inventory. Knowledge work doesn't, yet.
Read Full Note →The Dashboard That Lied
I built a dashboard that hit every number it was supposed to hit, and it took months to notice nobody's decisions actually changed because of it.
Read Full Note →Throughput Is Not Speed
If everyone works faster, does more value actually reach the customer? Not automatically — and that gap is the whole reason this research exists.
Read Full Note →Robot Adoption and AI Adoption
The robots in a 1984 novel made one resource look brilliant on a report and left the rest of the plant exactly as constrained as before. Read that as a period piece at your own risk.
Read Full Note →What Should AI Productivity Actually Measure?
Most AI adoption metrics measure activity: logins, prompts, hours theoretically saved. None of them ask whether more finished value reached a customer or a decision.
Read Full Note →Decision Inventory
How much of an organization's 'work' is actually just decisions waiting for someone to make them? Once I started looking for it, I found queues everywhere nobody called a queue.
Read Full Note →When Custom AI Beats Packaged Software
When is it actually worth building something instead of buying it? I default toward buying for most things and building for the handful of workflows where the fit really matters.
Read Full Note →Curiosity as an Operating System
What actually explains the difference between people who get real value from AI and people who don't? It isn't technical skill.
Read Full Note →People Don't Read Data. They Read What Their Brain Can Process.
The data was accurate, available, and regularly shared. It still wasn't changing behavior. The question wasn't whether the data was correct — it was why it wasn't landing.
Read Full Note →The AI Great Divide Isn't About AI
My husband sent me an article about a growing divide between two types of AI users. I laughed and said we were in the builder category. Then I decided I didn't agree there were only two.
Read Full Note →She Didn't Need the Answer
I expected my daughter to use AI to get answers. Instead, I watched her learn how to ask better ones. This isn't a rebuttal to the worry that AI makes people stop thinking — just a reason to hold it a little less absolutely.
Read Full Note →The Hidden Cost of Extra Hands
Every handoff had a good reason for existing once. I wanted to ask a different question: was this review actually adding value, or was it just delaying the work?
Read Full Note →