B2W AI Adoption Readiness

Four repos scored against Boris Cherny Steps of AI Adoption (0–4). Scope: repo harness readiness (verify, memory, isolation, gates) — not org seat count or team behavior. Audited 2026-08-04 from live codebases.

4 / 4

Repos at Step 1

0 / 4

Ready for Step 2

~25%

Avg Step 2 unlock

Verify + CI

Shared bottleneck

Org verdict

All four repos are Assisted-capable (pair programmer). None clear the Step 1→2 unlock: a trusted self-verification loop, worktree isolation, and automated review so agents can run in parallel without a human watching every keystroke.

Cherny steps (reference)

Step	Role	Agents	Unlock requires
0 Gated	—	0	Access path + secure launch
1 Assisted	Pair programmer	~1	Human reviews nearly every change
2 Parallel	Orchestrator	~10	Self-verify (tests/lint/build/e2e) + auto mode + worktrees + auto review
3 Supervised	Manager of managers	~100	Routines/loops, agent memory, sandboxing, trust in the loop
4 AI-native	Steer by intent	1,000+	Scaled domain automation + cost/quality guards

Source: Boris Cherny, Steps of AI Adoption (Jul 2026). Tokens alone do not advance a team — verification and guardrails do.

Step 2 unlock readiness by repo

taiyou-dw

30%

hime

28%

TEMP-app

22%

findTEMP-webapp

22%

Step 2 unlock (%) Rough Step 2 bar (70%)

Cross-repo scorecard

Repo	Domain	Step	→2 %	Agent docs	Verify	Isolation	Secrets
taiyou-dw	ClickHouse + Dagster DW	1	30%	Strong	Weak	Partial	OK
hime	Rails monolith + GraphQL	1	28%	Gap	Partial	Gap	Gap
TEMP-app	Expo / React Native	1	22%	Gap	Gap	Gap	Gap
findTEMP-webapp	React / Vite SPA	1	22%	Gap	Gap	Gap	Gap

Repos — open for detail

taiyou-dw Step 1 · 30%

Closest to Parallel — strong AGENTS.md

Blocked by missing CI / lint / typecheck and fragmented verify.

hime Step 1 · 28%

Rails + GraphQL — tests exist but thin

Blocked by deploy-only CI, tracked .env, no agent memory, fixed ports.

E2E API safety-net plan →

TEMP-app Step 1 · 22%

Expo RN — typed pair path

Blocked by zero tests, no CI, no AGENTS.md, Firebase/Sentry secrets in git.

findTEMP-webapp Step 1 · 22%

Vite SPA — lint/build ready

Blocked by zero tests, no CI, tracked .env*, README/codegen drift.

Level 2 Parallel roadmap →

Cross-cutting: what makes a task good enough to hand to an AI

Gates and tests are only half of Level 2. The other half is task quality — the 10-field spine, the four task types, and the role responsible for each field. Read the agent-ready task rubric →

Shared P0 to unlock Parallel

Priority	Action	Repos affected
P0	CI gate on every PR: lint + typecheck/build + tests	All 4
P0	Self-verify loop agents can run (tests/lint/build; e2e where UI)	All 4
P0	Agent memory: AGENTS.md / CLAUDE.md with verify commands + boundaries	hime, TEMP-app, findTEMP (taiyou-dw: extend)
P1	Secrets hygiene: untrack .env / Firebase / Sentry tokens; ship .env.example	hime, TEMP-app, findTEMP-webapp
P1	Worktree isolation (parameterized ports/DB/volumes)	All 4
P1	Default automated PR code + security review	All 4

B2W · AI adoption harness audit · Framework: Boris Cherny Steps of AI Adoption · 2026-08-04 · Share this folder (share/) — open index.html in a browser.