Capstone: Your Own 90-Day Plan

This is Unit Nine of Leading AI at Work, the capstone. Unit Eight closed on the question the method keeps asking: do we understand our things well enough to change them safely? This unit puts every earlier unit to work for one person, over ninety days. Carry one question through this lecture: how does one person's small change in practice grow into an organization's way of working? The answer arrives with the whole arc, at the very end. By the end you will be able to take the book's small challenge. To turn its result into a scoped pilot with an AI feature placed safely. And to connect that pilot to a playbook, an operating model, NIST's questions and a team that learns.

Four entries on the challenge the book ends with, and the whole method on a single page.

A challenge, not a pep talk.

LAI 201 The Small Challenge

The M-CaReM-AI method, by J. R. Carver, ends with a concrete challenge rather than a pep talk: pick one thing central in your world, a Case, a Client, an Incident, an Application, a Claim or an Order. Everything after that, the book says, is a deeper version of the same move.

A quick Silver Pack.

LAI 202 Drafting Your First Silver Pack

The challenge asks for a quick Silver Pack, two or three pages, not perfection. Write a definition and purpose a colleague would recognize. List ten to twenty key fields with their meanings. Sketch a simple lifecycle. List the cards you think you have. Write at least three open questions you already know about. Those questions are the most valuable part, because they mark where understanding runs out and the next conversation should start.

Show it, then use it.

LAI 203 Showing It and Using It Once

Two moves turn the draft into practice. Show it to one domain expert or teammate and ask, in the book's words, what is wrong, missing or confusing in your draft of what this thing means here. Treat the answers as validation and update the Pack. Then open it the next time you write a story, design a change, debug or prompt, and ask whether the work changes these things. Do this once, the book says, and you have taken your first real step.

The whole method, one page.

LAI 204 The Method on One Page

The Toolbox puts the method on one page: anchor requirements and AI work in a system's things and the changes to them. Mark the Gold ones, give core things Packs, describe behavior on cards, trace Gold things in matrices, and frame AI with Packs. AI acts on the system; people own what things are for.

Seven entries, constructed for this course, follow one analyst from her small challenge through every unit. The opening question is answered here.

Choosing a path and a scope.

LAI 205 Capstone One: Path and Scope

A capstone, constructed for this course, follows Priya Nair, a senior analyst at a regional water utility. Her small challenge was the Service Disconnection, the decision to shut off a household's water for unpaid bills. Showing her Silver Pack to a billing supervisor added a protected status for medical exemptions and an unwritten pending-review state. Using Unit One's steps, she proposes a single-project pilot on the new collections module, because the Disconnection's mistakes reach families, money and the law. Unit Two's test fits: real risk, a supportive manager, behind schedule but not on fire. She records the starting point: four regulator complaints last year from disconnections that should have been paused. The scope grew from evidence, not enthusiasm, which is what won the manager's yes.

The ninety days, planned.

LAI 206 Capstone Two: The Pilot Plan

The second part turns that scope into the book's ninety days. In the first fifteen, Priya names stakeholders from officers and compliance to the payment-risk model's engineer, and a noun storm yields twenty-six things, four zones, five core things and three Gold candidates: the Disconnection, the Payment Arrangement and the Disconnection Notice. By day forty, two Gold Packs and four cards exist, and joint reviews change the notice's timing rule. By day sixty, a Disconnection matrix includes the model as a row. The last thirty days put every change through it while she tracks complaints, rework and onboarding. Roles follow the book, with the manager owning purpose. The Arrangement waits for round two, in writing, so a deliberate delay is not mistaken for a forgotten thing.

A model, through the gate.

LAI 207 Capstone Three: An AI Feature Through the Gate

The third part takes the payment-risk model through the checklists and the gate. Its builder wants the score to set the order disconnections are processed. Priya writes its purpose, helping officers prioritize calls, and what it is not for: deciding a disconnection. It reads the Account, Payment History and Payment Arrangement, and influences the Disconnection, a Gold thing. The boundary test is clear: if nobody looks, a high score moves a household to the front of the disconnection queue, which is a decision. So the score orders the call list instead, and a disconnection still needs an officer's review of protected status. The safety checklist finds two gaps, missing field rules and missing tests for exempt households, both fixed before release. The model was unchanged; its place in the process moved.

A playbook that outlasts her.

LAI 208 Capstone Four: Playbook, Rituals, Metrics

The fourth part writes Priya's micro-playbook, since utilities are not among the book's six domains. Its Gold decision things are the Disconnection and the Arrangement, which customers dispute and the regulator audits; tracing statuses to the regulator's rules gives the medical exemption a clear basis. AI may draft notices, summarize histories and order call lists; it may never decide a disconnection or whether a household is protected. Then the operating model: Priya stewards the Disconnection, a billing writer edits Packs, the test lead is custodian, and an architect becomes AI Safety Lead with time set aside. The four rituals start, and she tracks time to clarity, rework and the model's override log. If she leaves, the hats, the review dates and the gate remain.

NIST's questions, answered.

LAI 209 Capstone Five: Govern and Manage

The fifth part has the utility's risk office ask how the model is governed. Priya answers from documents she already has. Govern one point six: the AI row of the Disconnection's matrix. Govern three point two: the gate record, where the model orders calls and an officer decides. Manage one point one: the written purpose. Manage one point two and one point three: unfair ordering of protected households, mitigated by the status check and tests. Manage one point four: a skimmed call list, recorded as residual risk. Manage two point four: the manager owns the switch back. Manage four point one: appeals and overrides feed the review. Asked about a simpler option under Manage two point one, she notes the rule-based call order kept as fallback. She wrote no new governance document; documents built for the work were the proof.

Teaching the team.

LAI 210 Capstone Six: Starting Your Team

The sixth part plans how Priya teaches the collections officers, who were outside the pilot. She starts with a ninety-minute clinic, because the Payment Arrangement is described three ways. Drafting its Pack with the officers, they find a missed payment has meant both a warning and an automatic cancellation, depending on who you ask. A month later, a one-day workshop for fourteen, including customer service and the model's engineer, keeps the Disconnection all day, and in the AI segment, on an approved tool with no personal data, every group catches a draft field for a customer's credit rating. Newer analysts get the self-study path; testers get the weekly drill. The clinic came first because people learn on the confusion they already have.

The whole arc.

LAI 211 The Whole Arc

The last part sets Priya's own ninety days beside the pilot. She keeps the four habits throughout. By day thirty, her portfolio holds two Packs, seven cards, a rough matrix and three prompts; by day ninety, four Packs, two Gold, fourteen cards, a working matrix, an owned prompt library, and notes on conversations the documents changed. She has helped two officers write first cards. When a regulator inquiry arrives about a disconnection that should have been paused, she answers in an afternoon from the Pack, the matrix and the review notes, and is asked to run a second pilot. In a year, those complaints fall from four to one, and the one was caught by the override log. That answers this lecture's question: one person working differently, with nobody needing to approve it, grew into a pilot, a placed AI feature, a playbook, an operating model, checkable governance and a team. To begin your own, go back to lesson one: one analyst, one Monday, and one thing.

Look at what you can now do. You can take the book's small challenge: pick one thing, draft a Silver Pack with its open questions, show it to someone, and use it once. You can turn that Pack into a scoped ninety-day pilot with named roles, and place an AI feature through the gate so it orders work around a decision instead of making it. And you can connect the pilot to a playbook, an operating model, NIST's govern and manage questions, and a team that learns on its own confusion. That is the course. Your next step is the small challenge.