Letters from the field

Practical writing. Pulled from the work.

Notes from active engagements — what we audit, what we ship, what we refuse, and the patterns we keep seeing across teams. Written by senior operators, edited for restraint.

Hand-drawn ink sketch of a magnifying lens held above a notebook page with a single brush stroke and Yale Blue underline

Audit8 min readApril 22, 2026

The AI readiness audit

What to check before you build.

Eight diagnostics we run before agreeing to any AI build. Half of them have nothing to do with AI.

Read

Vertical ink list of five abstract symbols, one circled in vermilion seal red with a Yale Blue arrow pointing right; the others struck through with hatch marks

Strategy7 min readApril 29, 2026

How to pick which workflow to automate first

ROI ranking, friction maps, and the test we use to refuse.

A scoring rubric for picking the workflow most likely to ship — and the four common mistakes that put the wrong workflow at the top.

Read

Hand-drawn ink system diagram with a single confident brush stroke struck diagonally across the upper third, leaving the lower portion intact; one arrow rendered in Yale Blue

Production9 min readMay 2, 2026

Why most AI POCs don't ship to production

The gap between demo-works and system-runs.

Four failure modes we see in every stalled POC, and the day-60 ship-cycle that closes them.

Read

Ink tally-mark grid arranged in five rows, several marks circled in vermilion seal and several crossed out, with a Yale Blue underscore on the bottom row

Governance10 min readMay 5, 2026

Eval harnesses and AI workflow governance

The eval discipline that makes a model swap routine.

What an eval harness is, when to build one, and the governance checklist you need before AI runs against production data.

Read

Three vertical ink columns of differing brush weights — solid, thin, dashed — with a short Yale Blue rule beneath the dashed column

Capacity8 min readMay 7, 2026

Build vs. hire vs. embed

Choosing your AI capacity model.

When in-house headcount makes sense, when consultants do, and when fractional embedding is the right shape. A decision matrix.

Read

Read the guide

These five articles are the guide. Read them in sequence.

The pillar page sequences them as a learning path — readiness audit first, capacity model last — with a short framing per cluster.

Open the guide