DevOps Agentic AI: 9.1 What "Agentic" Actually Means 86 9.1.1 A Spectrum, Not a Category 86 9.1.2 When Agency Genuinely Pays 87 9.2 Anatomy of the Agent Loop 87 9.3 Architectures 88 9.3.1 The Patterns 88 9.3.2 Choosing 89 9.4 Tools 89 9.4.1 Design Principles 89 9.4.2 The T

Learning Objectives

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

1. Define agentic systems precisely and locate a given design on the spectrum from fixed workflow to open-ended agent.
2. Select the simplest architecture that solves the problem and justify any move to greater autonomy.
3. Design tools that an agent can use reliably, including schemas, error semantics, and idempotency.
4. Architect agent memory across working context, session state, and persistent stores — and defend it against poisoning.
5. Engineer for reliability: compounding error, checkpointing, budgets, timeouts, and partial failure.
6. Evaluate agents on both outcome and trajectory and build a regression suite that survives non-determinism.
7. Operate agents in production: versioning, deployment, rollback, and observability.
8. Secure agents against prompt injection, confused-deputy attacks, and excessive agency.
9. Design human oversight that works, accounting for automation bias and the vigilance decrement.
10. Integrate agents with enterprise systems, including transactional and compensating behavior.