About AgentNotebook
AgentNotebook is an independent publication for engineers who build AI agents. We exist for one reason: most writing about agents stops at the idea. We do not. Every guide here is a working build you can reproduce, from the first empty file to a running agent.
What we publish
We publish step-by-step, runnable tutorials. You can copy the code, install the dependencies, and reach the same result we did. We do not publish vague overviews or roadmaps. We publish builds that compile, run, and do something useful by the end.
Each tutorial names its exact stack so you know what you are committing to before you start. That means a specific model provider (Anthropic Claude, OpenAI, or an open-source model), a specific runtime (Node or Python), and the supporting pieces around it: the Model Context Protocol, a vector store, a queue, or a deploy target. When a build touches paid APIs, we publish real cost figures: dollars per thousand requests, token pricing for the models used, and the monthly bill you should expect at a given volume. Cost is a first-class part of every architecture decision we document, not a footnote.
We organize tutorials around four stacks: building agents from scratch, adding agents to an existing SaaS, writing MCP servers, and evaluating agents so regressions surface before your users find them. We cover Anthropic, OpenAI, open-source models, and the wider MCP ecosystem. Our goal is to help you pick the tools that fit your stack, never to push a single platform.
Who writes AgentNotebook
AgentNotebook is founded and written by Ren Okabe and Sofia Nieves. Ren focuses on agent runtimes, tool loops, and the MCP ecosystem; Sofia focuses on evaluation, cost modeling, and shipping agents into production SaaS. Together they review every tutorial for technical accuracy before it goes out. When a guide ships code, we run it first. If a build later breaks because an API changed, we update the guide and note exactly what moved and when.
That review loop is the whole point. An agent tutorial that no longer runs is worse than no tutorial at all, so keeping these builds current is treated as ongoing work, not a one-time publish.
Contact
Found a bug in a tutorial, want a topic covered, or shipping something with these patterns? Use the contact form and we'll get back to you.