E
EVALS
Why your "90% accurate" LLM is failing in production
The gap between offline eval and live performance is almost always a coverage problem. A practical playbook.
Jun 2, 2026
Notebook
Field notes from the studio.
Writing for ML engineers, applied scientists, and the leaders deciding where to bet on AI next.
A
AGENTS
Tool use is the new prompt engineering
Frontier model quality has converged. The remaining alpha is in the tool surface you expose to the agent.
May 26, 2026
R
RAG
When to retrieve, when to fine-tune, when to do both
A decision tree from 40+ deployments. The defaults most teams pick are wrong about a third of the time.
May 19, 2026
C
COST
We cut our LLM bill 87% without losing quality
Five techniques: cascading, distillation, structured outputs, semantic caching, and ruthless prompt compression.
May 11, 2026
S
SAFETY
Red-teaming production agents: a starter battery
42 attack templates that catch real-world misuse before your users do.
May 2, 2026
G
GOVERNANCE
EU AI Act for US enterprises: what actually applies
Most of it does not. Here is what does, and how to map your existing GRC stack to it.
Apr 22, 2026