A free 52-chapter AI-native field guide

AI-Native Agent Engineering: AI Agent Development from Zero to Production

OpenAI Agents SDK · Claude Agent SDK · Production AI Harnesses

This complete AI agent development tutorial takes you from no prior agent knowledge to production expertise. You will build with the OpenAI Agents SDK and Claude Agent SDK, then engineer a vendor-neutral harness for tools, state, policy, budgets, approvals, traces, evaluations, security, recovery, and deployment.

Start from zeroExplore all 52 chapters
Chapters
52
Learning parts
6
SDK tracks
2
Capstones
3

Beginner to production engineer

What will this AI agent development tutorial teach you?

You will learn one connected engineering system rather than isolated demos. Every chapter upgrades SupportOps and adds evidence that the new capability works and fails safely.

  • 01

    Choose between deterministic code, workflows, single agents, and multi-agent systems

  • 02

    Build typed agents with tools, sessions, handoffs, subagents, MCP, and streaming

  • 03

    Design provider-neutral policy, state, budget, event, checkpoint, and adapter contracts

  • 04

    Evaluate quality, grounding, tool behavior, safety, latency, and cost independently

  • 05

    Secure tools, tenants, credentials, sandboxes, approvals, and external integrations

  • 06

    Deploy queued, resumable, observable agents with incident and migration runbooks

How does the book teach OpenAI and Claude without hiding their differences?

Shared foundations come first. Each SDK then receives its own accurate learning track, followed by adapters that normalize business lifecycle while preserving native features.

OpenAI Agents SDK

Explicit agent composition

Agents, typed output, function and hosted tools, sessions, handoffs, manager orchestration, guardrails, approval interruptions, MCP, traces, and evaluations.

Claude Agent SDK

Claude Code as a library

Built-in file and command tools, sessions, custom MCP tools, subagents, permissions, hooks, skills, plugins, cost tracking, OpenTelemetry, and secure hosting.

Complete curriculum

What is inside the 52-chapter AI agent engineering book?

The sequence moves from mental models to SDK implementation, then from harness contracts to production deployment, red teaming, operations, and migration. Every chapter has a direct answer, verification, failure exercise, knowledge check, and expert extension.

PART 01

Start Here

  1. 00About This Book
  2. 01What Is AI-Native Agent Engineering?
  3. 02How Does the Agent Loop Work?
  4. 03What Is an AI Harness?
PART 02

Foundations

  1. 01Python and TypeScript Essentials for Agent Builders
  2. 02Models, Messages, Instructions, and Context
  3. 03Structured Outputs and Validation
  4. 04Tools and Capability Boundaries
  5. 05State, Memory, and Sessions
  6. 06Retrieval, Grounding, and Citations
  7. 07How Do You Choose an Agent Architecture?
PART 03

OpenAI Agents SDK

  1. 01Set Up and Run Your First OpenAI Agent
  2. 02Define OpenAI Agents, Instructions, Models, and Output
  3. 03Add Function Tools and Hosted Tools
  4. 04Run, Stream, and Inspect OpenAI Agent Results
  5. 05Manage OpenAI Sessions and Resumable State
  6. 06Orchestrate OpenAI Agents with Handoffs and Managers
  7. 07Add Guardrails and Human Approval
  8. 08Connect OpenAI Agents with MCP and Integrations
  9. 09Trace, Debug, and Evaluate OpenAI Agent Workflows
  10. 10Capstone: Build SupportOps with the OpenAI Agents SDK
PART 04

Claude Agent SDK

  1. 01Set Up and Run Your First Claude Agent
  2. 02Understand the Claude Agent Loop and Built-In Tools
  3. 03Manage Claude Sessions, Resume, Fork, and Persistence
  4. 04Use Claude Structured Output and Streaming
  5. 05Give Claude Custom Tools with MCP
  6. 06Use Claude Subagents and Context Isolation
  7. 07Configure Claude Permissions, Approval, and Sandboxing
  8. 08Extend Claude with Hooks, Skills, Commands, Plugins, and Memory
  9. 09Track Claude Cost, Usage, and OpenTelemetry
  10. 10Host and Securely Deploy Claude Agents
  11. 11Capstone: Build a Claude Repository Maintenance Agent
PART 05

The AI Harness

  1. 01Design a Provider-Neutral AI Harness Contract
  2. 02Route Models and Providers with Evidence
  3. 03Engineer Context and Version Prompt Assets
  4. 04Build a Tool Registry and Policy Engine
  5. 05Build the State Store, Memory Lifecycle, and Compaction
  6. 06Engineer Queues, Retries, Idempotency, and Recovery
  7. 07Enforce Budgets, Stop Conditions, and Runaway Prevention
  8. 08Unify Events, Traces, Metrics, Logs, and Replay
  9. 09Build Evaluation and Regression Gates
  10. 10Secure Agents Against Injection, Leakage, and Cross-Tenant Access
  11. 11Design Human Oversight and Agent Incident Response
  12. 12Implement OpenAI and Claude Provider Adapters
PART 06

Production and Mastery

  1. 01Build the API, Worker, and Long-Running Agent Architecture
  2. 02Deploy with Containers, Sandboxes, and Secret Boundaries
  3. 03Engineer Reliability, Scaling, Latency, and Cost
  4. 04Choose Multi-Agent Patterns and Avoid Their Anti-Patterns
  5. 05Production Capstone: Build the Complete SupportOps Harness
  6. 06Red-Team and Evaluate the SupportOps Capstone
  7. 07Operate, Improve, and Migrate Agent Systems
  8. 08AI Agent Engineering Field Reference

Where should you go after the book?

Continue into the wider curriculum or apply the architecture to a real enterprise system with Muhammad Usman Akbar.

Build the wider engineering system

Continue with Digital FTEs Engineering, the AI-Native Azure book, and the AI-Native Sales playbook.

Design an enterprise agent platform

Explore AI-native consulting, learn how a Forward Deployed Engineer works, or book a strategy session.

Frequently asked questions

Is this AI agent development tutorial suitable for beginners?
Yes. The book assumes no previous agent knowledge. It begins with programming and language-model foundations, explains the agent loop and tools in plain English, and introduces production concepts only after the reader has observed and built the smaller behavior they depend on.
Does the book teach both OpenAI Agents SDK and Claude Agent SDK?
Yes. Separate hands-on tracks cover OpenAI agents, tools, handoffs, guardrails, sessions, tracing, and evaluations, plus Claude built-in tools, permissions, hooks, subagents, MCP, sessions, OpenTelemetry, hosting, and secure deployment in Python with focused TypeScript guidance.
What is an AI harness?
An AI harness is the control layer around an agent runtime. It owns trusted context, tool policy, permissions, budgets, durable state, checkpoints, approvals, telemetry, evaluations, recovery, and business outcomes so probabilistic model decisions can operate inside explicit production boundaries.
Will I build a real production-shaped project?
Yes. One SupportOps system grows throughout the book from typed classification to tool-using agents, OpenAI and Claude capstones, provider adapters, policy and state stores, queued workers, evaluations, red-team tests, isolated deployment, dashboards, approvals, and incident runbooks.
Does the book explain when not to use an AI agent?
Yes. The first architecture decision is whether deterministic code or a fixed workflow is safer and simpler. Multi-agent designs are introduced only when distinct ownership, context, tools, policy, models, or independent parallel work provide measurable value over the single-agent baseline.

Build the first agent—then engineer the system around it

Start with the mental model, follow both SDK tracks, and finish with the harness, production capstone, red-team suite, and field checklists.

Start reading freeDiscuss an agent platform