Description
Agentic AI: From Foundations to Frontiers explains the concept of intelligent systems that can act autonomously to achieve specific goals. It begins by covering the core foundations such as machine learning, natural language processing, and reinforcement learning. The book describes how AI agents perceive their environment, make decisions, and execute tasks efficiently. It also explains the architecture of agentic systems, including planning, memory, and tool integration. Various real-world applications in healthcare, education, and business are highlighted to show its practical impact. The discussion extends to advanced topics like multi-agent systems and collaborative AI environments. It emphasizes the growing role of large language models in enabling intelligent agents. Ethical considerations such as bias, safety, and accountability are also addressed. Additionally, the book explores challenges like reliability and high computational requirements. It concludes by presenting future directions where AI systems become more autonomous and transformative across industries.
About the Author
Contents
Preface v
1. Evolution of Artificial Intelligence and the Emergence of Agentic Systems 1
1.1 Introduction: From Intelligence to Execution 1
1.2 Phase I: The Analytics Era — Descriptive and Diagnostic Intelligence 3
1.3 Phase II: The Machine Learning Era — Predictive Intelligence 10
1.4 Phase III: The Large Language Model Era — Linguistic Intelligence 13
1.5 The Execution Gap: Why Enterprises Need More Than LLMs 16
1.6 Phase IV: Agentic AI — From Answers to Actions 24
1.7 Defining Agentic AI 25
1.8 Human–AI Collaboration as a Design Principle 27
1.9 Why Traditional Automation Fails Where Agentic AI Succeeds 29
1.10 Chapter Summary 31
2. Agents vs Chatbots and the CORE² Execution Model 35
2.1 Introduction: Why “Agents” Are Not Chatbots 35
2.2 Chatbots: Conversational Response Systems 37
2.3 Why Chatbots Fail in Enterprise Workflows 38
2.4 Agents: Goal-Directed Execution Systems 41
2.5 Agents Are Systems, Not Models 42
2.6 Chatbot vs Agent: A System-Level Comparison 43
2.7 Why Execution Requires a Behavioral Framework 44
2.8 The CORE² Execution Model: Overview 45
2.9 Context: Interpreting Intent and Constraints 47
2.10 Orchestrate: Planning and Coordination 47
2.11 Run: Deterministic Action Execution 48
2.12 Evaluate: Validation and Quality Control 48
2.13 Escalate: Human-in-the-Loop Governance 49
2.14 CORE² as a Closed-Loop System 50
2.15 Why CORE² Reduces Hallucinations 50
2.16 Agents as Accountable Digital Workers 51
2.17 Chapter Summary 55
3. Single-Agent Architectures: Reasoning, Tools, Memory, and Validation 59
3.1 Introduction: From Behavior to Architecture 59
3.2 The Single-Agent Architectural Stack 60
3.3 Context Interpreter 62
3.4 Reasoning and Orchestration Engine 65
3.5 Tool Execution Layer 68
3.6 Memory Subsystems 70
3.7 Validation and Evaluation Layer 73
3.8 Escalation and Governance Interface 75
3.9 End-to-End Flow of a Single Agent-A Complete Walkthrough 77
3.10 Common Failure Modes in Single-Agent Systems 82
3.11 Designing for Reliability 82
3.12 Chapter Summary 83
4. Multi-Agent Systems: Coordination, Roles, and Topologies 87
4.1 Introduction: Why One Agent Is Not Enough 87
4.2 What Is a Multi-Agent System? 88
4.3 Why MAS Are Essential in Enterprise Contexts 89
4.4 Canonical Agent Roles in MAS 91
4.5 Mapping CORE² to Multi-Agent Systems 94
4.6 Communication Between Agents 96
4.7 Coordination Topologies 96
4.8 Conflict Detection and Resolution 105
4.9 Synchronization and State Management 108
4.10 Failure Handling in MAS 110
4.11 Detailed Multi-Agent Workflow: Contract Review System 111
4.12 Design Principles for Reliable Multi-Agent Systems 116
4.13 Chapter Summary 117
5. Memory, Knowledge Grounding, and Retrieval-Augmented Generation 119
5.1 Introduction: Why Memory Matters more than Models 119
5.2 Memory as a First-Class System Component 120
5.3 Types of Memory in Agentic Systems 121
5.4 The Problem of Hallucination 123
5.5 Knowledge Grounding: The Core Principle 125
5.6 Retrieval-Augmented Generation (RAG) 126
5.7 Components of a RAG System 127
5.8 RAG Within the CORE² Model 131
5.9 Complete RAG Implementation: Retail Pricing Agent 132
5.10 Evaluation and Grounding Validation 136
5.11 Failure Modes in RAG Systems 136
5.12 Memory Governance and Safety 138
5.13 End-to-End Example: RAG in the Retail Pricing Agent 140
5.14 Designing Memory-Aware Agents: A Practical Framework 141
5.15 Chapter Summary 142
6. Tool Calling and Action Execution in Agentic AI 147
6.1 Introduction: From Knowledge to Action 147
6.2 Why Language Generation Cannot Execute Work 148
6.3 Defining Tools in Agentic Systems 149
6.4 Properties of Enterprise-Grade Tools 150
6.5 The Tool Execution Layer 152
6.6 Tool Selection and Authorization 157
6.7 Complete Tool Implementations: Retail Pricing Agent 159
6.8 Error Handling and Recovery 165
6.9 Tool Calling Within the CORE² Model 167
6.10 Tool Outputs as Evidence 168
6.11 Why Tool Calling Reduces Hallucinations 168
6.12 Security and Safety in Tool Execution 169
6.13 End-to-End: Tool Execution in the Retail Pricing Agent 170
6.14 Design Principles for Actionable Agents 172
6.15 Chapter Summary 173
7. Reasoning, Planning, and Autonomy Levels in Agentic AI 177
7.1 Introduction: Intelligence Is Not Autonomy 177
7.2 Reasoning in Agentic Systems 178
7.3 Why Reasoning Alone Is Insufficient 181
7.4 Planning as Structured Decision-Making 182
7.5 Planning Within CORE² 184
7.6 Autonomy as a Controlled Design Choice 185
7.7 The Five Autonomy Levels 186
7.8 Autonomy Is Not Uniform Across Actions 188
7.9 Confidence, Risk, and Autonomy 189
7.10 Escalation as an Autonomy Boundary 190
7.11 Failure Modes in Reasoning and Planning 191
7.12 Designing Responsible Autonomy: A Framework 193
7.13 Chapter Summary 193
8. Safety, Governance, and Explainability in Agentic AI 197
8.1 Introduction: Why Safety Is an Architectural Concern 197
8.2 The Risk Landscape of Agentic Systems 198
8.3 Governance as a System Design Principle 202
8.4 Policy Enforcement in CORE² 203
8.5 Safety Boundaries and Guardrails 204
8.6 Explainability: Beyond Model Interpretability 206
8.7 Three Forms of Explainability 206
8.8 Auditability and Traceability 208
8.9 Human-in-the-Loop Governance 209
8.10 Safety in Multi-Agent Systems 211
8.11 Common Safety Failures and Anti-Patterns 212
8.12 Designing Trustworthy Agentic Systems: A Checklist 215
8.13 Chapter Summary 216
9. Failure Modes, Evaluation, and Monitoring of Agentic Systems 221
9.1 Introduction: Failure Is Inevitable, Collapse Is Not 221
9.2 Taxonomy of Failures in Agentic AI 222
9.3 Common Failure Modes in Single-Agent Systems 226
9.4 Failure Modes in Multi-Agent Systems 228
9.5 Evaluating Agentic Systems: Beyond Accuracy 230
9.6 Task-Level Evaluation 231
9.7 Step-Level Evaluation 232
9.8 Confidence and Calibration 232
9.9 Monitoring Agentic Systems in Production 233
9.10 Incident Detection and Response 234
9.11 Stress Testing and Simulation 235
9.12 Continuous Improvement of Agentic Systems 237
9.13 Anti-Patterns in Evaluation and Monitoring 238
9.14 Designing for Resilience 239
9.15 Chapter Summary 239
10. Enterprise Case Studies and Applied Patterns in Agentic AI 243
10.1 Introduction: From Theory to Practice 243
10.2 Case Study 1: Document-Driven Compliance Review (Legal / Regulatory) 244
10.3 Case Study 2: Healthcare Information Synthesis (Clinical / Regulated) 249
10.4 Cross-Cutting Architectural Patterns 255
10.5 Choosing Autonomy Levels by Domain 258
10.6 Measuring Success in Enterprise Deployments 259
10.7 Lessons Learned from Enterprise Practice 260
10.11 Chapter Summary 261
Appendices 264
Reviews
There are no reviews yet.