ai-agents A field note by Vikrant Sharma
O'Reilly's new AI agents book is free and that matters
O'Reilly just published a 220-page guide to building AI agents, complete with code. It is free. That is not normal for a technical publisher of this size.
O’Reilly just published An Illustrated Guide to AI Agents, a 220-page book on building agents with LLMs. It is free to read online. That is not normal for a publisher this large. Most O’Reilly books sit behind a paywall or cost forty to sixty dollars. This one is public. No account wall, no trial period. You click the link and read. The timing is interesting. Every AI company is racing to ship agents right now. Anthropic has computer use, OpenAI has function calling, Google has Gemini tools. The techniques are moving faster than the documentation. A free, indexed resource from a known publisher fills a gap that blog posts and scattered docs do not. The book covers concepts and code. That split matters. Most agent tutorials skip the conceptual layer and jump straight to function calls. You end up with working code that you do not understand. This one explains the loop structure, the state management, the error recovery patterns before it shows you the implementation. It is also illustrated. Diagrams for agent flows, state transitions, tool interaction patterns. That is rarer than it should be in AI writing. Most technical AI content is walls of text or raw code blocks. Free technical books from established publishers usually mean one of two things. Either the topic is so hot that giving away content builds brand value, or the publisher is testing a new distribution model. O’Reilly has done both before. The first edition of the Python Machine Learning book was paywalled. The updated deep learning guides went free. This follows that pattern. The Hacker News post has four points and zero comments after almost three hours. That is low engagement for an O’Reilly release. Either the HN crowd already knows this material, or the headline does not make the free access obvious enough. The title says “guide” but not “free guide”. That matters for clickthrough. I will read the state management chapter. That is where most of my agent experiments fall apart. The initial tool call works, the follow-up call loses context, the loop does not know when to stop. If the book has a clean pattern for that, it is worth the hour.
Source: An Illustrated Guide to AI Agents:Concepts and Code for Building Agents with LLM