The Report That Came Back Shuffled
A parallel batch job reshuffling its output every run. Not an AI problem. A 30-year-old distributed systems problem living inside a modern pipeline.
Daily Log
17 articles - Daily dispatches — what I built, broke, and learned with AI today
-
-
Your AI Doesn't Remember You, and That's Most of the Problem
Swapping flat memory for semantic retrieval changed how my entire system operates. Here's the shape of the upgrade.
-
The Constraint Document Nobody Writes
Same pipeline, ten different niches. The differentiator isn't the code, it's a file most teams skip entirely.
-
The Constraint Document Nobody Writes
Same pipeline, ten different niches. The differentiator isn't the code, it's a file most teams skip entirely.
-
When 'Show Me' Accidentally Means 'Do It'
A two-word query triggered every scheduled job in the system. What that says about building AI that listens.
-
The Three-Step Gate Nobody Wants to Build
Splitting one task into three steps across sixteen properties, and the ceiling I haven't found yet.
-
Acknowledgment Theater and the Curation Problem
When your AI says 'got it' but didn't, and why curation is harder than generation.
-
You Don't Know What You've Built
An audit of your own systems reveals more than you expect.
-
The Day I Accidentally Automated My Own Job Description
Eight projects, one thousand interactions, and a creeping suspicion about what my job actually is now.
-
Seven Projects, One Saturday, Zero Context Collapse
When AI handles the switching costs, the ceiling on what counts as 'one day's work' disappears.
-
Why I Decided to Work in Public
When AI can produce a million words for pennies, the only scarce thing left is the truth behind them.
-
The Gap Between Capable and Aware
AI can do almost anything you ask. The hard part is teaching it what to ask itself.