Human-centered architecture
Integration is not attachment. It is architecture.
A tool placed on top of a fragmented process can accelerate fragmentation. A human-centered architecture begins by defining authority, responsibility, workflow, and feedback before AI is asked to scale anything.
The book’s Five-Floor Framework describes a practical ordering principle: authority flows downward from human purpose and responsibility flows upward from execution. AI can operate across layers of a system, but it cannot own the highest layer—the human meaning, accountability, and final judgment that give the system direction.
Build the human system first
Start by mapping the work as people actually experience it. Identify the decisions that create value, the people accountable for each decision, the information they need, and the moments when a choice must be slowed down rather than accelerated.
Then identify where AI may reduce friction: synthesis, retrieval, drafting, pattern detection, simulation, or routine execution. The use case should be explicit about what AI may do, what it may recommend, and what it may never decide on its own.
| Architecture layer | Human ownership | AI contribution |
|---|---|---|
| Purpose | Why the system exists and whom it serves. | None—this is a human leadership responsibility. |
| Standards | Policies, values, thresholds, and constraints. | Check work against declared standards. |
| Decisions | Judgment, trade-offs, and accountability. | Recommend, simulate, and summarize. |
| Workflow | Roles, handoffs, approvals, and escalation paths. | Automate approved steps and surface exceptions. |
| Execution | Deliver outcomes and observe impact. | Accelerate repeatable work and reporting. |
Design for exceptions, not only the happy path
A reliable architecture names its exception paths. What happens when an AI recommendation conflicts with human expertise? Who can override it? How is the override documented? What is the fallback when data is incomplete or a model cannot explain its result? These questions are the design of accountability.
- Name a decision owner for every high-impact AI-assisted workflow.
- Make escalation visible and easy to use.
- Separate advice from authority in the user experience.
- Review whether automation is changing incentives or hiding responsibility.
Further reading
These external resources provide context for the broader AI governance principles discussed here. They do not endorse HI → AI = IE.
Read the complete book or receive new framework notes by email.