Chapter 7. MLOps for Production-Ready AI and Agentic Systems
Over the past six chapters, you’ve built a comprehensive foundation: preparing data for GenAI applications (Chapter 2), constructing multimodal agents (Chapter 3), orchestrating agent teams (Chapter 4), establishing evaluation frameworks (Chapter 5), and optimizing models and infrastructure (Chapter 6). Each of these capabilities represents a critical pillar of what we call agent operations (AgentOps)—the systematic practices that transform working prototypes into production-ready systems.
Figure 7-1 maps these pillars across nine key dimensions. This chapter extends the pillars you’ve learned with production-specific practices while introducing three pillars essential for sustainable operations: observability, security and safety, and cost and capacity.
Figure 7-1. The nine pillars of AgentOps
The gap between “the model works” and “the model works in production” is wider for GenAI models than traditional ML. Identical prompts produce different outputs. Language evolves constantly. Agents maintain state across sessions. No single metric captures quality. Costs can explode through hidden operational overhead.
These challenges compound over time. Models that perform well at deployment gradually degrade as language patterns shift. Without proper versioning, teams can’t identify which model version is running or what data ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access