25. Production Readiness Index

Why does this matter?

You have completed the main course and understand the basics of agents. It is now time to move from a "learning agent" to a "production agent" that runs reliably, safely and efficiently in real environments.

This chapter is a prioritization guide and a quick reference for production topics. It helps you decide what to implement first and provides quick links to the detailed chapters.

A real-world case

Situation: You built a working agent locally. You want to deploy it to production but don't know where to start.

Problem:

• Too many production topics to learn at once
• You don't know what is critical vs. nice-to-have
• You need a prioritization guide

Solution: This index gives you a prioritized roadmap: start with the mandatory blocks (observability, cost control, security), then add topics as you grow.

Prioritization Guide

Mode 1: Urgent production launch in 1 day (minimal set)

If you need to get an agent into production right now, start with these three topics and one rule:

1. 19. Observability and Tracing — Without it you are blind. Needed immediately.
2. 20. Cost & Latency Engineering — Critical for budget control.
3. 17. Security and Governance — A mandatory production block.

And one rule — the context budget: the tools block is sent with every request and paid for by every request — 4 compact tools ≈ 742 tokens, and that is schemas only, before the system prompt and history. The source of truth on spending is PromptTokens from the response usage (ch. 03, ch. 13, ch. 20).

These three topics give you basic production readiness: you will see what is happening, control costs, and protect the system.

Mode 2: Planned hardening in 1–2 weeks (extended set)

If you have time for planned hardening, add topics as you grow:

Week 1:

• 21. Workflow and State Management — When agents run long tasks
• 22. Prompt and Program Management — When prompts change often
• 23. Evals in CI/CD — Automated quality checks

Week 2:

• 24. Data and Privacy — If you handle personal data

Overview of Production Topics

Mandatory production blocks (needed immediately)

19. Observability and Tracing

When needed: Immediately, as soon as the agent goes to production.

What's inside: Structured logging, tracing of agent runs and tool calls, metrics (latency, token usage, error rate), log correlation via run_id.

20. Cost & Latency Engineering

When needed: When the agent is used heavily or works with large contexts.

What's inside: Token budgets, iteration limits, caching, fallback models, timeouts.

17. Security and Governance

When needed: Immediately, as soon as the agent goes to production.

What's inside: Threat modeling for tool agents, tool categorization, protection against prompt injection, RBAC for tools, dry-run modes, auditing.

Topics to add as you grow

21. Workflow and State Management in Production

When needed: When agents run long tasks (minutes or hours), or you need idempotency or error handling with retry.

What's inside: Queues and asynchrony, scaling, distributed state. Basic concepts (idempotency, retries, deadlines) are covered in Chapter 11: State Management.

22. Prompt and Program Management

When needed: When prompts change often, there are multiple versions, or A/B testing is needed.

What's inside: Prompt versioning, prompt regressions via evals, configs and feature flags, A/B testing.

23. Evals in CI/CD

When needed: When prompts or code change often and automated quality checks are needed.

What's inside: Quality gates in CI/CD, threshold calibration against a baseline, cost/latency gates, live evals in production, dataset versioning, handling flaky cases, security tests.

Specialized topics

24. Data and Privacy

When needed: When the agent handles personal data (PII) or secrets.

What's inside: PII masking at the logging/tracing boundary and at tool output (the model's input is untouched), redaction by default, secret protection, vault at build time, storage with TTL, GDPR and correlation ID.

Note: Production aspects of RAG and Multi-Agent are covered in the base chapters 06 and 07. Model selection and decoding settings are covered in Chapter 01 and Lab 00.

Prioritization Algorithm

Don't try to learn everything at once. Use this algorithm:

1. Start with the mandatory production blocks:

   • Observability (logging, tracing) — needed immediately, without it you are blind
   • Cost & latency engineering — critical if the agent is used heavily
   • Security and Governance — a mandatory production block

2. Add topics as you grow:

   • Workflow/state — when agents run long tasks or idempotency is needed
   • Prompt/program management — when prompts change often or there are multiple versions
   • Evals in CI/CD — when automated quality checks are needed

3. Specialized topics:

   • Data/privacy — if you handle personal data
   • RAG/Multi-Agent production aspects — see the production notes in Chapter 06 and Chapter 07

Relation to Other Chapters

• Fundamentals: Basic concepts are covered in Chapters 01-12
• Advanced patterns: Ecosystem and patterns are covered in Chapters 13-18
• Production topics: Detailed production guides — chapters 19-24; this chapter is their navigator

What's Next?

After implementing production readiness, your agent is ready for deployment in the real world. Keep monitoring, iterating and improving based on production metrics and user feedback.

Appendix — use it as a set of checklists when bringing an agent to production: glossary, checklists, SOP templates and decision tables.