-
Why Retrieval Still Matters in the Age of Long Contexts
10 May 2026
Large context windows are impressive, but retrieval-augmented systems still win on cost, latency, and precision for most production use cases.
-
Evals Are Underrated (And Usually Wrong)
22 April 2026
Evaluation is the hardest part of building with LLMs. Most teams skip it, do it wrong, or build evals that measure the wrong thing entirely.
-
On Building Things That Ship
15 March 2026
Research taste and engineering taste are different skills. Getting good at both — without letting one kill the other — is the actual challenge.