Field Notes

Notes from the field.

Short essays on building, evaluating, and fielding AI systems. Hard-won lessons and the occasional strong opinion, straight from the founder's desk. No cadence, no fluff.

05 entries
Building · Evaluating · Fielding

Latest · Field noteThe cover

Small, adaptive models are underrated

The industry defaults to the largest model that fits the budget. My doctoral research pointed the other way, and the field is quietly catching up.

July 14, 2026 · 2 min readRead note

June 10, 2026·1 min read

Evaluation is where AI programs live or die

Everyone wants the model. Almost nobody budgets for proving it works. In high-assurance settings, that is exactly backwards.

Read note

May 5, 2026·1 min read

RAG earns its keep when the stakes are high

Retrieval-augmented generation is the most under-appreciated tool in the applied AI toolkit, if you build it with discipline.

Read note

April 2, 2026·2 min read

When a knowledge graph earns its keep

Knowledge graphs are not a default ingredient of applied AI. They are a deliberate choice, and there is a short list of signals that tell you when the investment pays off.

Read note

March 12, 2026·2 min read

An agent without an evaluation harness is a demo

Agentic AI raises the ceiling on what a system can do, and raises the bar for proving it. Autonomy is not a feature you add. It is a claim you have to earn.

Read note