DevOps AI Engineering: 8.1 Introduction 76 8.1.1 What AI Engineering Is 76 8.1.2 Where Enterprise AI Actually Fails 77 8.2 Choosing an Approach 78 8.2.1 Does This Need AI At All? 78 8.2.2 The Escalation Ladder 78 8.2.3 Build, Buy, or Host 79 8.3 Data Foundations 79 8.3.1 What A

Learning Objectives

By the end of this chapter, you will be able to:

1. Define AI engineering and explain why most enterprise AI failures are organizational rather than technical.
2. Select an approach — prompting, retrieval, fine-tuning, or training — justified by the problem rather than by novelty.
3. Design a data foundation with the lineage, quality, and access controls that AI systems require.
4. Apply MLOps practice: experiment tracking, model registry, feature consistency, drift detection, and retraining.
5. Build a retrieval-augmented generation system, including the access control and evaluation that separate a demo from a production system.
6. Construct an evaluation practice — the discipline that distinguishes AI engineering from prompt experimentation.
7. Operate inference systems with attention to latency, throughput, cost per request, and quality monitoring.
8. Assess AI-specific security risks, particularly prompt injection and excessive agency, and design assuming compromise.
9. Govern AI systems against emerging regulatory expectations, with documentation, human oversight, and risk classification.
10. Engineer agentic systems with bound authority and recognize where multi-agent architectures do and do not pay.