Preface

15

Objective of This Book

16

From Copilot to Agent

16

Target Audience

17

Structure of the Book

18

Acknowledgments

20

1 Agentic AI in the Enterprise

23

1.1 From Transactions to Intent: The Architectural Evolution of SAP

25

1.1.1 The Transactional Era: Monoliths, Transaction Codes, and the Transaction SE80 Paradigm

25

1.1.2 The User Experience Era: Decentralization, SAP Fiori Apps, and OData Gateways

26

1.1.3 The Agentic Era: The Paradigm of Generative Intention and the Agent Fabric

27

1.1.4 Architectural Comparison: Deterministic State Versus Stochastic Flow

29

1.2 The Agentic Shift: The App and Agent Paradigm

30

1.2.1 The Division of Labor: Determinism Versus Probability

30

1.2.2 The Mechanics of Coexistence: How an App and an Agent Collide

31

1.3 Industry Patterns: Art of the Possible Examples

33

1.3.1 Supply Chain Management: The Self-Healing Logistics Network

34

1.3.2 Finance: The Continuous, Zero-Touch Close

34

1.3.3 Spend Management: Autonomous Sourcing and Dynamic Negotiation

35

1.3.4 Human Capital Management: Contextual Talent Orchestration

36

1.3.5 Customer Experience: Proactive Retention and Resolution

36

1.4 Summary

37

2 Foundations of Agentic AI in SAP

39

2.1 The Agentic Loop from Prompt to Action

40

2.1.1 Perception (the Input)

41

2.1.2 Reasoning (the Brain)

42

2.1.3 Planning (the Strategy)

42

2.1.4 Action (the Hands)

43

2.1.5 Reflection (the Quality Check)

43

2.1.6 Walking the Loop: The Worked Example

44

2.2 Automation Versus Agency: A Fundamental Shift

45

2.2.1 The Spectrum: From Scripts to Reasoning

46

2.2.2 The Context Gap: Why Agents Succeed Where Bots Fail

48

2.2.3 The Hybrid Model: Agents as Orchestrators

49

2.2.4 The Decision Matrix: When to Use What?

49

2.3 The SAP Agentic Stack

50

2.3.1 Generative AI Hub: The Central Switchboard

53

2.3.2 SAP Domain Models: The Business Context Engine

56

2.3.3 SAP Knowledge Graph: The Semantic Grounding Layer

58

2.3.4 SAP Business Data Cloud: The Unified Data Foundation

62

2.4 Summary

65

3 SAP Enterprise Architecture for Agentic AI Systems

67

3.1 Shifting Enterprise Architecture Paradigm

68

3.2 The Role of the Architect

69

3.2.1 Augmented Enterprise Architect

70

3.2.2 Different Architect Personas

70

3.2.3 Why Agentic AI Changes the Enterprise Architecture Mandate

72

3.2.4 Expanded Responsibilities for Enterprise Architects in the Autonomous Enterprise

73

3.3 Capability Mapping

76

3.3.1 Business Capability Map

76

3.3.2 Technology Capability Model for AI

78

3.4 Data-Centric Design

82

3.5 AI Agent Hub in SAP LeanIX

85

3.5.1 Benefits

85

3.5.2 AI Agent Discovery

86

3.5.3 AI Agent Governance

86

3.5.4 Model Context Protocol Server for SAP LeanIX

89

3.6 Designing AI Agents for the Enterprise

90

3.6.1 Governing Principles

90

3.6.2 Design Principles

91

3.6.3 Operating Principles

92

3.7 Summary

93

4 Architectural Patterns for Business Processes

95

4.1 Decomposing Business Processes for Agency

96

4.1.1 The Agenticness Assessment Matrix

97

4.1.2 Deconstructing the Order-to-Cash Dispute

100

4.2 The Runtime Spectrum: Embedded, Managed, and Custom Agents

102

4.2.1 The Embedded Paradigm: Joule and Native Agenticness

103

4.2.2 The Managed Agent Runtime (Declarative Design)

105

4.2.3 The Custom Agent Runtime (Code-Based Orchestration)

107

4.2.4 Choosing a Runtime: Embedded, Managed, or Custom

109

4.3 Pattern 1: The Supervisor (Multiagent Orchestration)

110

4.3.1 Joule as the Central Supervisor

111

4.3.2 Designing the Tool Registry (Model Context Protocol)

112

4.3.3 Orchestration Styles for Multiagent Systems

114

4.4 Pattern 2: Human in the Loop

115

4.4.1 Redefining the Approval

116

4.4.2 The Approval Handoff Mechanism: From Routing to Reasoning

117

4.4.3 Designing for Exceptions: The Graceful Failover

119

4.5 Pattern 3: Reflection Loops (ReAct)

120

4.5.1 The ReAct Framework

121

4.5.2 The Architecture of Self-Correction: Moving Beyond Try/Catch

122

4.5.3 Bounding the Loop: The Architecture of Constraint

124

4.6 Summary

126

5 Reference Architecture for Agentic AI in the SAP Landscape

129

5.1 Component Architecture in the SAP Landscape

130

5.1.1 Data Layer

131

5.1.2 AI Layer

132

5.1.3 Decision Layer

133

5.1.4 User Experience Layer

134

5.1.5 Development Layer

134

5.1.6 Security Layer

135

5.2 How the Layers Work Together: An End-to-End Walkthrough

137

5.3 Architectural Principles for the Enterprise Architect

138

5.4 Summary

139

6 Building Low-Code Agents with Joule Studio

141

6.1 Introduction to Joule Studio

142

6.2 Getting Started with Joule Studio

145

6.2.1 Setup

145

6.2.2 Configuration

149

6.2.3 Developing AI Capabilities

154

6.2.4 Developing Joule Skills

154

6.2.5 Developing Joule Agents

159

6.3 Configuring Grounding

164

6.3.1 Configuring Document Grounding

164

6.3.2 Configuring an AI Agent

166

6.4 Testing and Monitoring

167

6.4.1 Testing

167

6.4.2 Deployment

169

6.5 Summary

171

7 Building Pro-Code Agents on Agent Fabric

173

7.1 The Pro-Code Pathway

173

7.2 Agentic Frameworks

175

7.2.1 LangGraph

175

7.2.2 Pydantic AI

190

7.2.3 Comparing LangGraph, Pydantic AI, and n8n

198

7.3 Joule Code-Based Agent

200

7.3.1 Component Architecture

200

7.3.2 Setting Up the Joule Code-Based Agent

202

7.3.3 Understanding the Joule Code-Based Agent Framework

208

7.3.4 Testing and Deployment

211

7.3.5 Observability, Debugging, and Tracing

212

7.4 Integration with Agent-to-Agent

213

7.4.1 Architecture Overview

214

7.4.2 Invoking a Third-Party Agent via Agent-to-Agent

216

7.4.3 Security Considerations for Agent-to-Agent

217

7.5 Agent Harness for Production

221

7.5.1 Scope and Boundary

221

7.5.2 Context Engineering

222

7.5.3 Security Architecture

224

7.5.4 Deployment Architecture

225

7.6 Summary

226

8 Building Agents with Vibe Coding

227

8.1 Joule in the Autonomous Enterprise

228

8.1.1 Joule: From Digital Assistant to Engagement Layer

228

8.1.2 SAP-Managed Joule

230

8.1.3 Architectural Path

231

8.2 The Intent-Driven Workflow

233

8.3 Headless Architecture

233

8.4 Joule Studio 2.0

237

8.5 From Vibe to Pro-Code: A Seamless Graduation

238

8.5.1 One Artifact, Three Paths: Vibe, Low-Code, and Pro-Code

238

8.5.2 Agent Development

239

8.5.3 App Development

241

8.5.4 Workflow/Process Flow Development Using n8n

247

8.5.5 Comparing Low-Code, Pro-Code, and Vibe Coding

254

8.6 Common Pitfalls of Vibe Coding

255

8.7 Summary

256

9 Security, Trust, and Governance

257

9.1 Introduction: Why Autonomous Agents Require New Security Models

258

9.2 The SAP Framework for Responsible AI

259

9.2.1 Pillar 1: AI Ethics and Safety

260

9.2.2 Pillar 2: AI Security and Data Privacy

261

9.2.3 Pillar 3: AI Compliance

261

9.3 The SAP AI Trust Layer

262

9.3.1 Why a Raw Large Language Model Cannot Be Deployed Directly

263

9.3.2 Architectural Positioning

264

9.3.3 Input Filtering

265

9.3.4 Output Filtering

267

9.3.5 From Trust Layer to Agent Hub

268

9.4 Threat Modeling for Autonomous Agents

268

9.4.1 Chain-of-Thought Hijacking

269

9.4.2 Tool Use Vulnerabilities

270

9.4.3 Memory Manipulation

270

9.4.4 How to Implement Threat Modeling in SAP

271

9.5 Enterprise Security Architecture: Identity and Access in the Agentic Era

283

9.5.1 The Privilege Escalation Trap (the God Mode Problem)

284

9.5.2 Principal Propagation: The Identity Lifeline

285

9.5.3 How to Configure Principal Propagation for an Agent Identity

286

9.5.4 Resource Groups: The Isolation Boundary

290

9.5.5 Joule Sovereignty: Mapping Autonomy to the Human

291

9.6 Auditable Reasoning

291

9.6.1 The End of Code-Only Audits

292

9.6.2 Three Audit Artifacts That Didn’t Exist Before

292

9.6.3 The Evidence Chain: What the Platform Produces and What You Must Build

293

9.6.4 Why This Matters for the Chief Information Security Officer and the Audit Committee

294

9.6.5 How to Establish Auditable Reasoning Logs for Compliance

294

9.7 Setting Up an Agent Governance Board

302

9.7.1 Mandate and Charter

303

9.7.2 Composition: Four Core Roles

304

9.7.3 The Agent Intake and Review Process

304

9.7.4 Ongoing Oversight Cadence

305

9.7.5 Integration with SAP AI Agent Hub

306

9.7.6 Aggregate Trust Telemetry Dashboard: Weekly Review

311

9.7.7 Override and Intervention Panel: Real Time

313

9.8 Summary

314

10 Operating the Agent Ecosystem

317

10.1 The AgentOps Mandate: Moving from Pilot to Scale

318

10.1.1 The Pilot Purgatory (the 74 Percent Problem)

318

10.1.2 The Fallacy of Outside-In AI

319

10.1.3 The AgentOps Paradigm: Managing Stochastic Systems

320

10.1.4 The Agent-Led Toolchain

322

10.2 The Generative AI Center of Excellence

324

10.2.1 The Mandate: From Shadow AI to Governed Innovation

325

10.2.2 Roles in the Agentic Workforce

327

10.2.3 The Generative AI Center of Excellence as the Transformation Engine

329

10.3 Agent Discovery and the Registry of Intent: SAP LeanIX

331

10.3.1 The Risk of Registry Rot

332

10.3.2 The Agentic Register: Mapping Intent

332

10.3.3 Tool-Calling Lineage and Impact Analysis

335

10.3.4 Continuous Rationalization of the Synthetic Workforce

336

10.4 Technical Lifecycle and Observability: SAP Cloud ALM

338

10.4.1 Agentic DevOps and Continuous Integration/Continuous Delivery

338

10.4.2 Token Metering and FinOps

340

10.4.3 Deep-Loop Observability (OpenTelemetry)

342

10.4.4 The Runtime Kill Switch Protocol

343

10.5 Business Process Monitoring: SAP Signavio

344

10.5.1 Process Mining the Un-User Experience

345

10.5.2 The Key Performance Indicators of Autonomy: Measuring Core Impact

346

10.5.3 Value-Based Governance and the Feedback Loop

349

10.5.4 Validating Agent-Led Transformation

350

10.6 User Adoption and Enablement

351

10.6.1 The Trust Paradox

351

10.6.2 Confidence Scores as the New User Interface

352

10.6.3 From Operator to Orchestrator (the New Human Advantage)

353

10.6.4 Vibe Coding as a Cultural Catalyst

355

10.7 Summary

356

11 The Agentic Roadmap and Future Outlook

359

11.1 The Autonomous Enterprise

360

11.1.1 Why Now: Value Is Shifting to the Agentic Layer

360

11.1.2 SAP’s Right to Win

361

11.1.3 The Foundation: SAP Business AI Platform

362

11.1.4 The Execution Paradigm and the Clean Core Prerequisite

363

11.1.5 The Five Autonomous Domains

364

11.1.6 Real-Time Coordination and Built-In Trust

366

11.2 Beyond Joule: The Orchestration Era

367

11.2.1 Joule Work: The Evolution of the Enterprise Workspace

368

11.2.2 Assistants and Agents: Teammates and Executors

369

11.2.3 SAP’s Agent Fabric: Orchestrating the Fleet

370

11.2.4 Cognitive Capabilities: Planning, Reflection, and Cross-System Reasoning

370

11.3 Decentralized Ecosystems: Agent to Agent and Model Context Protocol

372

11.3.1 Model Context Protocol: Unifying the Tooling Ecosystem

372

11.3.2 Agent-to-Agent Protocols: Business-to-Business Cross-Boundary Negotiation

373

11.3.3 The Open Partner Agent Ecosystem

374

11.4 SAP and Hyperscalers

375

11.4.1 Model Sovereignty via the Generative AI Hub

376

11.4.2 Zero Data Retention and Collaborative Innovation

377

11.4.3 Cloud Portability and Future Proofing

378

11.5 Conclusion: The Beginning of Better

379

11.5.1 What Changes: The Transformation of Enterprise Operations

379

11.5.2 What Doesn’t Change: The Immutable Laws of Enterprise Architecture

379

11.5.3 The Path Forward for the Enterprise Architect

380

11.5.4 Key Takeaways for the Enterprise Architect

380