Learn and build
The open-source shelf for agent builders
26 free GitHub repositories we recommend for learning AI and building agents, arranged in the same order as the book: learn the foundations, build, connect tools, evaluate, then run and govern.
Pairs with Part 1 · Understanding agents
Learn the foundations
Free, structured courses that take you from first prompt to first agent.
- microsoft/ai-agents-for-beginners(opens GitHub in a new tab)A free course of short lessons on agent patterns, tool use, planning and multi-agent design.The clearest step-by-step route in for non-specialists.
- huggingface/agents-course(opens GitHub in a new tab)Hugging Face’s free agents course, with hands-on units and a certificate.Good for learning what happens inside the agent loop.
- microsoft/generative-ai-for-beginners(opens GitHub in a new tab)A broad lesson series on generative AI, from prompts to retrieval and safety.Fills in the basics before you get to agents.
- dair-ai/Prompt-Engineering-Guide(opens GitHub in a new tab)A guide to prompting techniques, with papers and examples.Instructions are part of the agent contract. Write them well.
Pairs with Part 7 · Worked examples
Worked examples from the model labs
Notebooks from the model providers, showing patterns that already work.
- anthropics/claude-cookbooks(opens GitHub in a new tab)Notebooks and recipes for tool use, retrieval, sub-agents and evaluation.Includes agent patterns that match the book’s bounded-task approach.
- openai/openai-cookbook(opens GitHub in a new tab)Examples and guides for common tasks with the OpenAI API.A wide library of patterns to borrow and adapt.
Pairs with Part 4 · Building and governing
Frameworks for building agents
Pick one, learn it well, and keep the agent’s boundary in your own design, not the framework’s.
- langchain-ai/langgraph(opens GitHub in a new tab)Build agents as explicit graphs of steps, with state, checkpoints and human-in-the-loop.Makes control points and approvals visible in the code.
- openai/openai-agents-python(opens GitHub in a new tab)A lightweight Python SDK for agents, handoffs, guardrails and tracing.Small surface area, with guardrails and tracing built in.
- google/adk-python(opens GitHub in a new tab)Google’s Agent Development Kit for building, evaluating and deploying agents.Treats evaluation as part of the build, not an afterthought.
- pydantic/pydantic-ai(opens GitHub in a new tab)A Python agent framework built around typed, validated inputs and outputs.Typed outputs make agent decisions easier to check.
- microsoft/autogen(opens GitHub in a new tab)A framework for multi-agent systems where agents work together on a task.Useful when one bounded agent really does need to hand off to another.
- crewAIInc/crewAI(opens GitHub in a new tab)Orchestrates role-based teams of agents with tasks and processes.A quick way to prototype work split across several roles.
- huggingface/smolagents(opens GitHub in a new tab)A minimal library for agents that think in code.Small enough to read end to end and understand.
- vercel/ai(opens GitHub in a new tab)A TypeScript toolkit for AI apps and agents, with tool calling and streaming.The quickest route to an agent with a web front end.
- run-llama/llama_index(opens GitHub in a new tab)A framework for connecting agents to your documents and data.The evidence layer: grounding decisions in the right sources.
Pairs with Chapter 8 · Controls and autonomy
Tools and connections
Every tool is a permission. Use these to connect an agent to the systems it is allowed to touch.
- modelcontextprotocol/modelcontextprotocol(opens GitHub in a new tab)The specification and docs for the Model Context Protocol (MCP).A standard way to give an agent tools with clear boundaries.
- modelcontextprotocol/servers(opens GitHub in a new tab)Reference MCP servers for files, Git, databases, search and more.Ready-made, inspectable tools to start from.
Pairs with Part 5 · Evaluation
Evaluation
If you cannot evaluate the work, you do not yet control it. These turn a golden set into a release decision.
- promptfoo/promptfoo(opens GitHub in a new tab)Test prompts, agents and retrieval against test cases, and red-team them, from the command line or CI.Puts evals in the release pipeline, where they can block a bad change.
- confident-ai/deepeval(opens GitHub in a new tab)A unit-testing style framework for LLM outputs, with ready-made metrics.Familiar to engineers who already write tests.
- openai/evals(opens GitHub in a new tab)A framework and registry for evaluating model behaviour.Shows how to structure evals you can repeat.
Pairs with Part 6 · Running agents in the organisation
Guardrails and observability
Most of an agent’s life happens after go-live. See what it did, and stop what it should not do.
- langfuse/langfuse(opens GitHub in a new tab)Open-source tracing, evals, prompt management and cost tracking for LLM apps.Cost per good outcome needs traces and costs in one place.
- NVIDIA-NeMo/Guardrails(opens GitHub in a new tab)A toolkit for adding programmable guardrails to conversational systems.Puts rules for what the agent may say and do into code.
- guardrails-ai/guardrails(opens GitHub in a new tab)Validates and corrects model inputs and outputs against rules you define.A control layer you can test separately from the model.
For inspiration
Collections to browse
Curated lists and example apps. Use them for ideas, then run each idea through the five-gate test.
- NirDiamant/GenAI_Agents(opens GitHub in a new tab)Tutorials and notebooks for agent techniques, from simple to advanced.Many patterns, each explained on its own.
- Shubhamsaboo/awesome-llm-apps(opens GitHub in a new tab)A large collection of example LLM apps and agents with runnable code.Concrete examples to pull apart and learn from.
- e2b-dev/awesome-ai-agents(opens GitHub in a new tab)A curated list of AI agent projects, products and frameworks.A map of the landscape, to see what already exists.
We are not affiliated with any of these projects. Open-source projects change quickly, so check each repository's licence, maintenance activity and security posture before you use it in production.
Get the book
The best agent is not the most autonomous one.
It is the one whose value, boundary, controls and evidence are clear.
Available now
Kindle eBook
Read tonight on any device. Searchable, with every canvas and checklist one tap away.
Buy on AmazonOut 20 November 2026
Paperback
The desk copy for design reviews, with the workbook canvases ready to mark up.
Paperback out 20 November 2026For organisations
Team copies and talks
Bulk copies for your team, or a talk or workshop on applying the method to your own agents.
Message Anuj on LinkedIn