Notes from the bench.
Working notes on building decision systems — what shipped, what got cut, and what we would do differently.
01On the rail
2 notes
All notes
The table nobody owns
Every stalled AI project we have inherited had the same root cause, and it was never the model. It was a dependency on data that no one was accountable for.
The baseline is your current process, not a test set
Held-out accuracy tells you how well a model fits history. It says almost nothing about whether the system will be worth running on Monday.
Tell us about one decision.
One decision your team makes often, and what it costs when it goes wrong.
Start a pilot