What I'm going to write here

I'm starting to write here weekly in September. This is a placeholder so the section, the feed, and the formatting all work before the first real post lands.

What this will be about

The gap between an AI demo and an AI system that survives inside a regulated organization — evaluation, reliability, and the unglamorous work that decides whether a pilot ships or dies at month two. Concretely:

• Agent reliability — making non-deterministic systems trustworthy enough to sign off on.
• Evaluation — how you actually know an LLM system works, beyond a vibe check.
• Forward-deployed notes — what breaks when the customer, the data, and the compliance team enter the room.

Formatting works

Posts support MDX, so I can drop in code:

def is_shippable(system) -> bool:
    # month two is the real test, not the demo
    return system.is_observable and system.is_reproducible

...and images, tables, and callouts. See you in September.