Introduction: From Chatbot to Coworker
There is a gap between what most people think AI can do and what it can actually do right now.
Most people are still using AI like a search engine with better grammar: ask a question, read the answer, move on. That's useful. But it's table stakes in 2026. The people pulling serious leverage from AI aren't just chatting with it — they're deploying it. They're running it overnight on tasks that used to take days. They're building agents that can open files, search the web, write and execute code, and report back with finished work.
This book is about how to do exactly that, using Anthropic's Claude Managed Agents.
Who This Is For
You do not need to be a developer to read this book. About half the people who will get value from it are marketers, coaches, consultants, and business owners who have some familiarity with AI tools but have never touched an API. The other half are developers, technical founders, and product managers who want to move fast without building agent infrastructure from scratch.
Both groups will find what they need here. Non-technical readers will come away with a clear mental model of how agents work, what to look for when hiring a developer to build one, and how to prototype ideas in Anthropic's Console before writing a single line of code. Technical readers will find the API details, code snippets, and architectural reasoning to build production-ready agents confidently.
What You Will Build
By the end of this book, you will understand how to:
- Create and configure a Claude agent with the right model, tools, and system prompt for your use case
- Run that agent in a managed cloud environment — no server setup, no Docker, no devops overhead
- Send it work, watch it think in real time through the event stream, and steer it when it goes off course
- Give it memory so it carries knowledge from one session to the next
- Control exactly which actions it can take automatically versus which need your approval first
These aren't theoretical concepts. Every chapter ends with a step-by-step walkthrough you can execute today, using the official ant CLI or the Python SDK.
How This Book Is Structured
Part 1 — Foundations builds the mental model. What is an agent, really? How does it differ from a chatbot? What are the four building blocks every Managed Agent is made of?
Part 2 — Build Your First Agent gets you to a working agent. You will configure its name, model, system prompt, and tools. You will learn what the eight built-in tools do and when to use them. You will set up permission policies so your agent acts confidently on low-risk tasks and asks before taking consequential ones.
Part 3 — Run Your Agent covers sessions and the event stream: how a session differs from a chat conversation, how to watch your agent think and act live, and how to interrupt or redirect it mid-task.
Part 4 — Give Your Agent Memory covers the feature that transforms a capable tool into something closer to a genuine coworker: persistent memory across sessions. You will learn the six memory tools, the limits that matter, and the patterns that work in the real world.
A Note on Sources
Every technical claim in this book is cited inline to official Anthropic documentation. Code snippets are reproduced verbatim from the official docs. Where the official docs were silent or ambiguous at time of writing, this book says so honestly rather than speculating.
Claude Managed Agents launched in public beta on April 8, 2026. This book reflects the API as it existed at that launch. Some features — including Memory, Multi-agent sessions, and Outcomes — are in Research Preview and require separate access. Each chapter notes clearly when a feature has this status.
The transformation this book promises is real and achievable. Let's get into it.