Discipline 04

Design

What system, at what complexity, does the workflow need?

Transformation eventually becomes a design problem. Strategy says what to change; architecture says what to build. Programs stall less often because the strategy is wrong than because there is nothing coherent to build toward: local decisions, vendor point solutions, and AI components with no operating logic holding them together.

Design is the discipline that prevents that. It starts from the systems the enterprise already runs and asks how AI connects them, because the hard problem is integration, not intelligence. It treats the capability stack as a capital allocation framework, so money reaches the data, knowledge and control layers that visible applications depend on. It specifies the control plane that enforces policy at runtime, and the team functions that build and run it.

Above all, it matches complexity to the workflow. An assistive copilot, a bounded automation, an autonomous agent and a regulated human-in-the-loop decision each warrant a different level of control. The shift to agents raises the stakes: an agent holds delegated authority, and the runtime around the model, with its identity, policy and protocols, becomes the platform.

Read01

Capability Stack

The seven-layer enterprise AI capability stack as a capital allocation framework: what each layer does, who owns it, and where to build or buy.

7 min read
02

Systems Model

How AI fits across the four systems every enterprise already runs, record, engagement, intelligence and action, and where value leaks between them.

6 min read
03

Control Architecture

The technical control plane that enforces governance at runtime across seven domains, from agent identity and entitlements to override and observability.

7 min read
04

Operating Architecture

The five team functions that build and run enterprise AI, who owns which decision, and how broad the platform team should be.

6 min read
05

Reference Patterns

Four deployment patterns, assistive, workflow automation, agentic and regulated human-in-loop, with the control, governance and maturity each requires.

6 min read
06

The Agentic Shift

Why agentic AI is delegated authority rather than a software feature, and what that changes in governance, cost, metrics and the human role.

6 min read
07

Protocol Landscape

How MCP, A2A, AIP and LDP compose into a multi-agent stack that is vendor-portable, auditable and interoperable, and in what order to adopt them.

8 min read
ProofProof

Decision Artifacts

Worked board memo, investment scorecard, model inventory, phase-gate review and risk classification worksheet, each with a downloadable template.

10 min read
Next discipline05 OperateIs it working in production, and do the economics hold?