Insights

Notes from systems in the field.

Practical writing on agent design, automation reliability, evaluation, and the operational side of applied AI.

01

Engineering note

Postmortem: the 2 AM financial-news pipeline

What failed, what changed, and how the production workflow became observable and recoverable.

02

Working thesis

Where AI belongs in an automation system

Use models for ambiguity; use schemas, constraints, and deterministic code for everything else.

03

Field note

Designing for the human reviewer

Agreement scores are useful, but evidence, controls, and correction paths make an editorial agent operational.

LLM-ready/index.md
Download .md