Blog
Writing on LLM execution, runtime reliability, and deterministic AI systems.
Why would you build the enforcement layer yourself?
The instinct to build the enforcement layer yourself is good engineering. Here is what the complete picture looks like, from someone who built it.
·7 min read
Why the model can't be its own compliance check
The model can't verify its own outputs. The check has to be external. Here's what the enforcement layer owns and what changes when it exists.
·6 min read
Why your LLM forgets your constraints mid-workflow
...and why you can't prompt your way out of it.
·6 min read
The math nobody does before shipping multi-step LLM workflows
The model did not get dumber. The failure compounded.
·6 min read