# Adoption playbook — from one person to the org

Fluency doesn't arrive by mandate. It compounds from a small, measured habit into
a team capability. This is the path: prove it on yourself, pilot it on a team,
measure honestly, then scale what worked.

## Phase 0 — Individual on-ramp (30 days)

The whole of Tier 1, run for real:

1. Pick **one** repetitive task you already do.
2. Do it **with AI for a month**.
3. Keep a **one-line note every time it fails you** (`lessons-log.md`).
4. Stand up a `CLAUDE.md` for the work so the AI stops repeating mistakes.

**Exit when:** you have a populated lessons log, a feel for where it breaks, and a
defensible before/after on time or quality. That evidence is what earns the pilot.

## Phase 1 — Team pilot (one squad, one quarter)

- Choose **one squad and one real workflow** — not a toy.
- Set the controls up front: a budget (`cost-controls.md`), a routing policy
  (`model-tiers.md`), and an audit trail (`audit-trail.md`).
- Define success **before you start**: the metric, the baseline, the threshold.

## Phase 2 — Measure (honestly)

Hold the pilot to the standard the talk argues for — verify with data, don't
trust the green check:

- **Cost per successful outcome**, vs the pre-AI baseline.
- **Quality** — defects, rework, review findings — not just speed.
- **Where it gamed the metric** — the failures are the most valuable output.

A pilot that honestly reports "not worth it here" is a success: it saved the org
a costly rollout.

## Phase 3 — Scale what worked

- Promote the squad's best lessons into shared `CLAUDE.md` invariants and policy.
- Make the controls org defaults: budgets, tiering, audit trail on by default.
- Grow person-by-person via the 30-day on-ramp — not a big-bang mandate.
- **Re-baseline quarterly.** Models, prices and your own workflows all move.

## The arc in one line

> Individual fluency earns a measured pilot; an honest pilot earns the rollout.
> Spend follows evidence, not enthusiasm.