$ LS ~/BLOG

Writing.
Notes from shipping AI agents into real companies.

On agent evaluation, context engineering, AI security, and the economics of running agents in production. Each post links to the full original.

Autonomy Isn't a Model Property. It's a Line in Your Config.

How much AI agents act without human oversight is a deployment decision, not a property of the model.

Teams rolling out AI coding agents are burning through annual budgets in months. Here's how to forecast it.

Agentic planning moves EPM from reports people wait for to answers people ask for.

Why single-run benchmarks overstate agent reliability, and what to measure instead.

A decision trace behind every agent action turns a black box into a glass box.

Keyword filters can't stop prompt injection. Enforce deterministic rules before the model.

Why specialized models and owned infrastructure hold up better than thin wrappers.