An attempt to define an enterprise AI adoption strategy · a working version

AI adoption, stage by stage

AI (artificial intelligence) adoption in a regulated organization does not start with agents. It starts with productivity inside guardrails: an AI inventory and intake, approved tools, shared team context, and guardrails as code. Each role climbs the adoption ladder at its own pace, and every use case carries its own risk tier. It is written for the hardest case, a regulated organization; any organization can follow the same order with lighter controls.

At the top of the ladder, delivery looks like ask-to-release: artifacts drafted up front, agents doing bounded work, and humans holding the decisions they are accountable for.

Explore the worked example →

A working version, not a final standard. Every organization has its own workflow, controls and people, so no single strategy fits all of them. What carries over is the shape: the stages, the risk tiers, the decisions that stay human, and what never scales down. Take the shape, then tune each table to your own workflow.

This is not a case for slowing down, and it is not a case for any vendor. The claim is that governance done first is what lets a regulated team move faster: when the inventory, the risk tiers and the guardrails exist, each new use case is a short intake, not a new negotiation with risk.

The thesis

Most regulated organizations buy a tool, hand out seats, and wait for value. Usage rises, then flattens. The pilots look good, but none of them reach production, because risk sees them for the first time at the end.

This site describes the other order. Governance first: an inventory, an intake, risk tiers and approved configurations, so that every later step has a known path. Context second: the team's instructions file, reusable prompts and an asset catalog, so the AI works with what the organization already has. Delegation last: bounded tasks behind evaluation and approval gates, then agentic workflows where the controls are proven.

Role tracks show the same path for nineteen roles, from the first steps to the fully adopted workflow, and how a small or unregulated organization covers each one.

The running example on every page is one ask at a regulated consumer-finance company: show the contractual payment due date the same way on web, mobile, statements and reminder notifications, and show the payment cutoff time in the customer's own time zone.

Two axes, not one

Adoption has two independent dimensions. Confusing them is the most common mistake: a team at a high stage does not earn the right to automate a high-risk use case.

Axis	Measured per	Values	Who decides	What it controls
Adoption maturity	Role, per team	Stage 0 Foundations → 1 Personal productivity → 2 Team context → 3 Delegated tasks → 4 Agentic workflows → 5 Scale and sustain	The team, with evidence the enablement lead checks	How much the AI does and what must exist first
Use-case risk tier	Use case, at intake	Low · Medium · High · Prohibited	Intake triage, with risk, privacy and security partners	Which controls the use case needs, at any stage

The two axes meet in one rule: a stage is not an entitlement to autonomy. A high-risk use case may stay human-operated at Stage 4. A prohibited use case stays prohibited at every stage. Roles move at different speeds, so QA (quality assurance) at Stage 3 while the product owner is at Stage 1 is normal.

The reuse story, in three lines

The organization had ten different date-and-time implementations (illustrative), although a shared library already existed.
Three of them treated the date-only due date as an instant, so West-coast customers saw their due date one day early.
Reuse first is not tidiness: it is a domain-correctness control, and it can be written as code: an advisory agent hook plus CI (continuous integration) checks. See chapter 7.

The model on one page

Read left to right. Governance creates the paths, each role climbs its own stages, reuse keeps the work grounded in what exists, and every change lands as an artifact a human approves.

How to read this site

Click any box in a diagram. It opens what the box is and what it is in the due-date example.
? marks a glossary term. Governance, adoption, agentic and delivery terms open in place; the full list is chapter 14.
Abbreviations are expanded under every diagram. In prose they are spelled out the first time they appear on a page.
Numbers are illustrative. Counts, durations and percentages show the shape of the model, not a benchmark.

Chapters

Chapter 1Why AI adoption stallsSix patterns that keep teams stuck between a pilot and real use, and the ten date implementations that show the cost. Chapter 2Governance is the first enablerThree lines of defense, two axes (maturity and risk), risk tiers, and a control crosswalk from the controls you already have to the AI failure modes they must now cover. Chapter 3Who decides, who runs itDecision rights, the use-case lifecycle from intake to retirement, exceptions, incidents and the evidence system of record. Chapter 4Finding the right use casesDiscovery, value × feasibility × risk scoring, the limits of AI, and when not to use it. Chapter 5The adoption ladderStages 0 to 5: what AI does at each, what must exist first, the evidence to move up, and when to step down. Chapter 6Role tracksNineteen roles, from the CIO to production support and frontline operations: what each decides, how it adopts AI stage by stage, and the evidence it owns. Chapter 7Reuse first, as codeA governed asset catalog, advisory agent hooks, a reuse decision record, and CI checks that enforce it. Chapter 8The worked exampleOne ask, the contractual due date and the payment cutoff, from Stage 0 to Stage 4, every role, every artifact, and the paths that go wrong. Chapter 9The enablement programFrom awareness to adoption: learning paths, workshops, labs, champions, the coaching cycle and change management. Chapter 10Measuring adoption honestlyA balanced metric tree: adoption, value, quality, risk and sustainability, with definitions and anti-gaming notes. Chapter 11Evaluating and mapping toolsCapability first: how to evaluate AI tools, and how the same concepts appear in today's coding agents. Chapter 12Hands-on labsSix labs from a team instructions file to a policy assistant that cites and abstains. Chapter 13Templates, each filledCopy-ready artifacts, blank and filled from the worked example. Chapter 14GlossaryEvery governance, adoption, agentic and delivery term on this site, with drill-down. AboutAbout this siteWhat this is, who wrote it, and what it is not.

What this is not

Guidance and patterns, not legal or regulatory advice. Scope and terms, such as what counts as a "model", follow each organization's own policy and jurisdiction.

Related sites: ask-to-release (what Stage 4 delivery looks like) · claude.sammuthu.com (architect study guide) · scaling.sammuthu.com (scaling agents).