All Pillars
2

ARCHITECT

What foundations do we need?

Build the factory before building the product. Organizations that scale AI successfully invest in reusable infrastructure — orchestration layers, data pipelines, identity frameworks, shared building blocks — before deploying individual agents.

12 controlsExecutive Sponsor: CIO or CTO

Assessment Controls (12)

Every AI initiative that passes through this pillar must satisfy these controls. The maturity model measures how consistently the organization enforces them.

1

Data Pipeline Governance

Are data pipelines governed and accessible to AI tools through controlled interfaces?

2

AI Agent Identity Management

Do AI agents have their own machine identities, or do they inherit human user credentials?

3

Multi-Agent Orchestration

Is there an orchestration layer for coordinating multi-agent workflows?

4

Reusable AI Components

Are there reusable building blocks (templates, connectors, shared services) across AI initiatives?

5

Agent Development Velocity

Can the organization spin up a new agent initiative in weeks (factory model) or does each one take months (bespoke model)?

6

Data Quality & Bias Assessment

Is the data infrastructure audited for quality, completeness, and bias before agents access it?

7

Third-Party AI Integration Inventory

Are third-party integrations (APIs, plugins, external tools) inventoried and assessed?

8

Agent Development Lifecycle

Is there a standard agent development lifecycle (design, test, deploy, monitor)?

9

Prompt Gateway & DLP Enforcement

Is there a centralized prompt gateway that enforces DLP, PII masking, and injection filtering before prompts reach any LLM?

10

RAG Pipeline Security

Are RAG pipelines secured? (Embedding encryption, RBAC on vector queries, retrieval-layer injection monitoring)

11

Measurement Mechanism Design

Is the mechanism for measuring outcomes designed during architecture, not bolted on after deployment?

12

Enablement Plan Design

Are training and adoption plans built during the architecture phase? Who needs to be trained, on what, by when — defined before rollout, not after.

Governance Tracks

Employee Use [EU]: What identity framework governs employee AI tool access?
Internal Build [IB]: Is there a standard architecture for internal agent development?
Vendor Platform [VP]: How do vendor AI features integrate with existing data infrastructure?