Field notes / 001—

Notes from the
production edge.

Practical observations on inference, AI platforms and the engineering work between a model demo and a dependable product.

The benchmark said one second. Production said ninety.

What theoretical load tests reveal—and why KV-cache locality decides whether an agent feels instant or broken.

Next
Platform

Keys and routing are the easy part. Ownership, fallbacks and observability decide whether teams trust the platform.

Drafting

Next
Reliability

Why leases, idempotency and atomic artifacts matter when inference lasts minutes instead of milliseconds.

Drafting