AIAF architecture views
Status: These views describe delegated, proportionate decisions and the ten-course case story. The diagrams describe the SEI Artificial Intelligence Adoption Framework (AIAF); they do not authorize any particular organization's use of AI.
Use the framework overview for definitions, the capability model for capability-record and level rules, the decision workbook for the case record, and the lifecycle decision map for the phase-to-decision interpretation. The diagrams group the 63 capabilities and ten output types for legibility; they are not substitutes for those sources.
Architectural context: what the framework connects
Text equivalent: An accountable sponsor and case team define an initiative and its adoption case around sustained business value, with trust as a required condition. Affected people inform the objective and consequences. The case is classified by what AI does (one or more W1–W4 workloads, with W4 authority where applicable) and how AI is obtained and controlled (delivery mode A–D). The case moves through nine lifecycle phases. Seven capability perspectives and the 63-record catalog provide applicable organizational outcomes; applicability and four levels describe responsibility and required reliability, not extra lifecycle phases. The team applies the eight-step assessment method as needed, retaining evidence, decisions, dependencies, measures, and plans among ten required output types. Observed outcomes and changed conditions feed the next decision. Providers and specialists contribute evidence and work, but do not assume the organization's accountability. Five coordinated products teach, record, communicate, and maintain this model. Distinctive position, guardrails, and review questions constrain its use.
The overview's framework promise, central objective, framework at a glance (including the initiative/case distinction and relationships), workloads, delivery modes, lifecycle, capability perspectives, capability levels, assessment method, required outputs, product architecture, distinctive position, guardrails, and review questions all appear in this context view. Individual capability records, workbook fields, and the detailed contents of each output belong in the linked reference documents, not a context diagram.
Sequence: how a case is assessed and moved
Text equivalent: The sponsor names the objective; the team opens an adoption case. At each lifecycle phase or material change, the team identifies the next useful action, uses the framework's classification and capability guidance, gathers context-specific evidence from relevant parties, and applies the relevant portions of the eight-step method. It updates the case record rather than starting ten new documents each time. An accountable decision-maker chooses a bounded path and records its conditions, owner, and review trigger. That person may be a team member acting within delegated authority; a wider decision-maker or specialist is involved when spend, exposure, obligations, or consequences exceed that authority. The team acts, observes outcomes, and re-enters the decision cycle. The framework does not treat provider claims as accepted organizational evidence automatically.
The five named Investigate, Select, Pilot, Expand, and Continue checkpoints are landmarks inside this loop; the lifecycle decision map says what they are and are not, and makes design, evaluation readiness, routine operation, and exit closure visible. The formality of each decision follows the organization's delegated limits and the consequence of the proposed action.
Conceptual information model: what persists
Text equivalent: One enduring business initiative may contain multiple narrower adoption cases. Each case has a classification, current lifecycle phase, applicable capability profile, and evidence register. Evidence supports a bounded accountable decision; capability gaps can constrain it. The decision permits progress or sends the case back and directs roadmap actions. Those actions produce new evidence. The initiative's objective persists even if a particular AI solution, provider, or case is changed or retired.
How the views work together
- Coherent core: The initiative/adoption-case distinction, independent workload and delivery-mode classifications, cross-cutting capabilities, proportionate levels, evidence-based decisions, and feedback into a roadmap are compatible. The framework has a credible organizing model rather than five unrelated lists.
- Decision architecture clarified: Nine phases describe the case's work, and the five named checkpoints are landmarks within them (see the lifecycle decision map). The map supplies the other decision functions, including the distinction between pilot expansion and acceptance into sustained operation. The workbook retains this distinction without adding mandatory approval boards.
- Repeatable assessment: The eight assessment steps are applied as needed, while the ten output types form an evolving case record. The sequence and conceptual views show how evidence and decisions are revisited.
- Current course-pathway status: Ten released courses now show evaluation, observed pilot outcomes, operation, and verified retirement as distinct learner performances. The Pathway Companion brings the case decisions and roadmap together across the series. See the ten-course audit.
- Documentation/product distinction: The five intended products are ways to deliver and maintain the same framework. They are not lifecycle phases, steps, output types, or governance authorities.
Method, output, and course-coverage check
This is a coverage diagnostic, not a proposed new course sequence. One output can be revised by several method steps and at several lifecycle phases.
| Assessment-method step | Principal output types updated | Existing or planned teaching coverage |
|---|---|---|
| 1. Define objective | Objective/value record; initial measures and boundaries | Course 1 teaches the initial case and investigation request. Later phases must refine baselines, cost, and outcomes. |
| 2. Classify scenario | Workload/authority/delivery classification; consequence record | Courses 2–3 teach classification and consequence, including provisional choices that must be revisited when the proposed use changes. |
| 3. Determine applicability | Consequence/applicability decisions; capability profile | Course 4 teaches selective capability applicability. |
| 4. Set target levels | Current/target capability profile | Course 4 teaches proportionate targets for a narrow decision, not a whole-organization maturity score. |
| 5. Assess evidence | Evidence/assumption register; measures and review triggers; provider acceptance | Evidence quality is introduced in Course 1 and applied to discovery, design, evaluation, pilot, operation, and material change in Courses 5–10. |
| 6. Identify gaps and relationships | Capability gaps; shared-responsibility/dependency records; continuity/exit conditions | Course 4 identifies critical gaps; Courses 6–7 test people/provider dependencies; Courses 9–10 test operating and exit dependencies. |
| 7. Build roadmap | Prioritized roadmap; measures, owners, and dependencies | Each decision card updates next actions and review triggers. The Pathway Companion brings the case roadmap together across the ten courses. |
| 8. Review outcomes | Realized measures, revised decisions, roadmap and exit conditions | Course 8 reviews the bounded pilot, Course 9 reviews first routine-operation evidence, and Course 10 reviews several later periods plus retirement closure. |
The ten-course pathway supplies those lifecycle observations and decisions without making any one course equal one phase. Learners can use the evolving record and roadmap to explain the decision chain from a different organizational starting point. The course count remains a consequence of distinct workplace performances, not of the eight-step, ten-output, nine-phase, or five-checkpoint counts.