Essays on building data infrastructure for AI agents, the difference between precision and prescription in LLM product design, and what I've learned from shipping three major projects in a year.
-
If you prescribe every outcome, you've built in every bias you already have. Precision and prescription aren't the same thing. Most teams are conflating them.
-
When an agent starts behaving unreliably, the instinct is to add more instructions. It makes things worse. The fix is precision, not volume.
-
One camp engineers the agent so thoroughly no human has to intervene. The other believes the human-AI collaboration is the product. I'm in camp two.
-
I don't assume it knows everything, and I don't try to make it do everything. I use it as a thinking partner. Here's what that actually looks like.
-
I didn't start in data. The skill set from process safety management consulting transferred more directly than you'd expect: understand a complex system, find the failure points, build something that holds up under pressure.
