AI Onboarding Sprint · Sample
Cadence — four weeks to a team running AI without us
Not trained on AI — running it, on their own work, with the guardrails built in.
PREPARED WITH CADENCE PRODUCT TEAM (SYNTHETIC) · 6 PEOPLE · 4-WEEK SPRINT + 30-DAY CHECK-IN · OUSIOS LLC
Day-30 adoption
Stuck running unaided, guardrail live
At risk used, but drifting — tune-up
Not yet installed, not adopted
Why a sprint, not a workshopThe gap this closes
Cadence's six-person product team didn't need another AI course. AI adoption fails for a behavioral reason, not a technical one: workshops teach features, the team learns to prompt, but how the work gets done never changes — and the judgment guardrails (where AI quietly fails at each task) get skipped. So we didn't train. Over four weeks we installed four workflows into the team's real, recurring work, each with its guardrail as a team standard, and judged success by one thing: are they still running, unaided, at day 30?
The outcome — judged at day 30, up top by designWhat actually stuck
Day-30 verdict
Three of four workflows are running unaided with their guardrails live as team norms. One is at risk: the weekly landscape scan is being run, but shortcuts around its source-verification guardrail — flagged, with a tune-up booked, not called a win.
The honest read: the team no longer needs us for three of these; the fourth needs one more nudge before we can say it's a habit.
Adoption: 3 STUCK · 1 AT-RISK · 0 NOT-YET — we report what stuck, not what we hoped would.
What we installedFour workflows on real weekly work — with the guardrail as the standard
| Workflow | On whose real work | The guardrail, now a team standard | Day 30 |
|---|
| Research Synthesis | PM's weekly user-interview digest | Every claim tagged known / assumed — no synthesis ships as fact unverified. | Stuck |
| Decision Brief | Eng lead's recurring build-vs-defer calls | ≥3 options incl. do-nothing, plus the strongest case against — no one-answer memos. | Stuck |
| Weekly Status Report | Team lead's Monday leadership update | Auto-draft, but a human names the one thing that isn't moving — the signal the draft flattens. | Stuck |
| Landscape Scan | PMM's competitor watch | Every factual claim carries a VERIFY tag until a human opens the source. | At risk |
We capped it at four (the range is 3–5). A fifth was on the shortlist — the vision-artifact workflow — and cut: the team doesn't do that work weekly, so it wouldn't have stuck. Impressive-but-unused is the failure mode.
The four weeksAssess → install → hand off & sustain
- Week 1 · Assess & selectComposa profiles + pick the right 3–5 Working-style profiles set who leads which workflow and where adoption friction sits. We chose workflows the team does weekly — not impressive ones they'd never reuse. (Profiles guide fit; never a competence ranking.)
- Weeks 2–3 · Install with the teamRun it on real work, build the runbook as we go Each workflow installed on live, recurring work — the team runs it, we coach — with a one-page runbook per workflow and its guardrail baked in as the standard.
- Week 4 · Hand off & sustainSustainability plan + schedule the 30-day check-in Owners, triggers, guardrails, and a "when this breaks, here's what to check" page. The engagement isn't closed until the day-30 check-in is on the calendar.
- Day 30 · Check-in & tune-upJudge adoption honestly, fix what's drifting Return to measure what stuck. Three running unaided; one tuned up. Outcome judged here, not at week 4.
The sustainability planWhat keeps these running after we leave
- An owner per workflow — the person whose real work it lives on, named, not a shared "the team."
- A trigger per workflow — the recurring moment it fires (Monday update, every user-interview batch), so it's a habit, not a decision.
- The guardrail as the standard — each Composa caveat written into the team's norms, not left as a footnote.
- A "when this breaks" page — the two failure modes per workflow and what to check first. (The Anchor's reliability layer.)
- The 30-day check-in — on the calendar before we left. The loop only closes if drift gets caught.
The honest part — flagged, not hiddenWhat could still make this not stick
At risk at day 30 — named, with a fix, not called a win
A vendor optimizing for usage would report "4 workflows adopted!" We won't — one is drifting, and hiding it would cost the team the trust the whole sprint is built on.
- Landscape Scan is skipping its source-verification guardrail: the PMM runs it, but pastes unverified claims into competitor briefs. Adoption without the judgment guardrail is the failure mode. Tune-up booked: re-anchor the VERIFY habit and make it a review-gate, not a suggestion.
- Owner-dependence risk on the Decision Brief: only the eng lead runs it. If they're out, does it survive? Flagged to cross-train a second owner before it's called a durable habit.
The at-risk flags are part of the deliverable. "Adoption that survives our exit" is the product — and you can't manage drift you pretend isn't there.
The discipline, made visibleWhy this isn't an AI training workshop
| AI training workshop / course | This sprint |
|---|
| Teaches features and prompting | Installs workflows into real, recurring work |
| Judgment guardrails skipped | Each guardrail becomes a team standard |
| Knowledge fades within a month | Owners, triggers, and a "when it breaks" page |
| Success = attendance | Success = adoption measured at day 30 |
| Confidently-wrong output, eroded trust | Drift flagged honestly, then tuned |
Adoption without judgment is the failure mode. The product is a team that doesn't need us anymore.
Want your team running AI — not just trained on it?
Four weeks, one team, 3–5 workflows installed on real work with the guardrails built in, and a 30-day check-in to prove it stuck. Fixed scope, one department budget. Book an Onboarding Sprint →