The Archive

Essays & Long-form

Thorough examinations of the systems we are building and the structures of evaluation, governance, and trust that we are not.

Stage

Topic

MAY 2026

Why You Can't Serve LLMs Like Regular Models

A practical guide to the five fundamental differences between traditional ML inference and modern LLM serving — continuous batching, prefill/decode disaggregation, PagedAttention, prefix-aware routing, and MoE sharding.

Evergreen AI INFRASTRUCTURE 25 min

Earlier writing (2020) is preserved in the repository but no longer surfaced here. New essays land in this list as they're published.