Chapters 07–10

Humans and agents, under clear rules

How roles change, who decides what, and how governance scales autonomy without becoming the brake.

07. Human + agent operating model

Humans Don't Disappear. Their Role Changes.

Traditional model

Humans execute the work.

Human
Task→Task→Decision→Task→Outcome

Agentic model

Agents execute routine and coordinated work. Humans direct it.

Agents
ExecuteMonitorCoordinateAdapt
Humans
Set goalsExercise judgmentHandle exceptionsProvide oversightRemain accountable

From human execution to human orchestration and oversight.

Different processes need different levels of autonomy

Select a level
Less autonomyGovernance required ↑More autonomy

Decision rights for every agent

Autonomy only works when everyone knows who decides. Settle these five questions before an agent goes live.

QuestionOwnerRule
Who approves deployment?Business ownerNo agent goes live without a named business owner, after review by the agent owner and risk.
Who monitors its behavior?Agent ownerPerformance, drift and incidents are tracked continuously against agreed thresholds.
Who can intervene or stop it?Process ownerEvery agent has a human override and a stop switch that works in minutes, not weeks.
Who owns the handoff between teams?End-to-end process ownerAccountability gets lost in the transitions. One person owns the whole flow.
Who is accountable for the outcome?Business unit leaderOwning the model is not the same as owning the decision it influences.

Final accountability stays singular and human.

Autonomy needs architecture.

08. Governed autonomy

More Autonomy Requires Better Governance

Enterprise agentic AI cannot scale safely through uncontrolled agent proliferation. Every agent needs an identity, boundaries and someone accountable for it.

Agents across the enterprise

Service agentProcurement agentFinance agentPlanning agentResearch agent

Governance control layer

Agent RegistryIdentity & AccessPermissionsObservabilityEvaluationAudit LogsHuman OversightEscalation RulesRisk ClassificationCompliance Controls

Enterprise systems, data and people

ERPCRMData platformKnowledge baseProcess owners
Low risk
Greater agent autonomy
Medium risk
Bounded autonomy with monitoring
High risk
Human approval before action
Critical decisions
Human decision, supported by AI
Exhibit 3 · Autonomy follows riskBar = agent autonomy

The goal is not maximum autonomy. The goal is the right autonomy for the right risk, without governance becoming the brake itself.

  • ProportionalLight controls for low-risk agents, deep controls for critical decisions.
  • Paved roadsApproved, reusable patterns replace case-by-case approval.
  • PrunedA control that no longer reduces risk is removed, not kept by habit.
09. The shift

From AI Projects to AI Transformation

FromTo
Scattered AI pilotsStrategic AI programs
Individual use casesEnd-to-end business processes
AI toolsAgentic workflows
Siloed AI teamsCross-functional transformation squads
Human executionHuman-agent collaboration
ExperimentationIndustrialized delivery
Technology metricsBusiness outcomes
AI governance documentsGoverned autonomy
AI adoptionEnterprise transformation
10. The enterprise foundation

Four Capabilities Make Agentic Transformation Scalable

People

  • AI literacy
  • Workforce redesign
  • Human-agent collaboration
  • New roles
  • Adoption
  • Change management
  • Leadership

Governance

  • Agent autonomy
  • Accountability
  • Risk classification
  • Human oversight
  • Monitoring
  • Compliance
  • Escalation

Technology

  • Agent orchestration
  • Enterprise integration
  • APIs
  • Observability
  • Modular architecture
  • Interoperability
  • Vendor flexibility

Data & Knowledge

  • Accessible enterprise data
  • Reusable data products
  • Data quality
  • Knowledge architecture
  • Unstructured data governance
  • Permissions
  • Trusted context
Scalable agentic operating model
NextApproach & toolsFrom first step to scale