Best AI Agents Repos, Guides, and Videos for 2026

If you have finished the courses and read the papers, the next move is to get your hands on real code. This is the list of AI agents repos to clone, the two engineering guides that actually explain how to ship, and the talks that walk through an agent loop end to end. It is the open-source shelf that turns theory into something you can run, fork, and break. Treat it as the working bench for the 12-post Learn AI Agents series: where the series teaches the concepts, these 18 picks are the materials you build with.

The list is organized into three groups by what each resource is for. Repos are the codebases worth cloning for runnable patterns, from teaching curricula to reference architectures. Guides are the two practitioner essays that tell you how production agents are really built. Videos are the talks that show an agent being assembled on screen, which is often faster to absorb than a README. Together they are the practical companion to the 12-post Learn AI Agents series on this site.

Repos: the AI agents repos worth cloning in 2026

These are the AI agents repos where the value is in the code, not the README. Some are full curricula you follow lesson by lesson, others are reference implementations you read to see how a working engineer structured a real system. Star counts are live from GitHub at the time of writing.

• GenAI Agents (nirdiamant). The pattern library most agent tutorials quietly borrow from, with a folder per pattern, reflection, planning, tool use, RAG, multi-agent, each as a runnable notebook rather than a slide. Around 23.6k stars is a lot for a teaching repo, and it stays relevant because the author keeps refreshing it as new patterns land. Clone it when you want to see the skeleton of an agent technique before you reach for a heavy framework. Open GenAI Agents on GitHub.
• Hands-On Large Language Models (HandsOnLLM). The official code for the Tunstall, von Werra, Werry book, covering everything from tokenization and fine-tuning to in-context learning and evaluation. Around 27.9k stars reflects how many readers treat it as the de facto LLM cookbook, and it pairs cleanly with the agents material because every agent assumes the LLM fundamentals in these notebooks. Start here if you are shaky on how a model actually produces a token before you build a loop around it. Open Hands-On Large Language Models on GitHub.
• Made with ML (GokuMohandas). One of the most loved applied-ML curricula on the site, at roughly 48.9k stars, and still the clearest bridge from notebooks to production, with lessons on experimentation, packaging, and deployment that most agent tutorials skip. It is the repo to clone when you realize your agent needs monitoring, versioning, and a deployment story, not just a prompt. Think of it as the engineering layer that sits underneath a fleet of agents. Open Made with ML on GitHub.
• Hands-On AI Engineering (Sumanth077). A growing roadmap and resource index for the AI engineer role specifically, smaller at around 2.8k stars but more focused on the current stack than the older curricula. It collects links and notes on LLMs, RAG, agents, vector stores, and evaluation in one navigable tree, which makes it a good bookmark when you are scoping what to learn next. Use it as a syllabus cross-check rather than a codebase to copy from. Open Hands-On AI Engineering on GitHub.
• Awesome Generative AI Guide (aishwaryanr). A curated, regularly updated list of generative AI papers, tools, and tutorials, organized by topic and date, sitting at about 28.5k stars. It earns its keep as a discovery layer, the place you check what landed this month before digging into a specific repo, and the maintainer’s monthly summaries are a genuine time-saver. Treat it as the index that sits on top of the more focused repos in this list. Open Awesome Generative AI Guide on GitHub.
• Designing Machine Learning Systems (chiphuyen). The companion repo to Chip Huyen’s book, with references, slides, and reading notes on ML systems design, at around 5.1k stars. It is the slower, more architectural read on this list, the one you reach for when you need to reason about requirements, data, metrics, and deployment trade-offs before writing a line of agent code. Pair it with Made with ML: one for the how, one for the why. Open Designing Machine Learning Systems on GitHub.
• Machine Learning for Beginners from Microsoft (Microsoft). A 12-week, 26-lesson curriculum that takes a learner from zero to a working grasp of classical ML, and at about 88.9k stars it is the most-starred teaching repo on this list. It is here because every agent engineer eventually hits a wall where the problem is not the LLM but the underlying model, the features, or the evaluation, and this is the gentlest on-ramp to fix that. Skim it before you assume a bigger prompt will solve your accuracy problem. Open Machine Learning for Beginners from Microsoft on GitHub.
• AWS Guidance for Enterprise Agentic AI Platform (Mehdi Nemlaghi and Cristian Balint / AWS Solutions Library). Where the rest of this list teaches patterns or curates links, this one ships an opinionated, deploy-ready reference platform for running AI agents on Amazon Bedrock AgentCore at enterprise scale. It is a new, AWS-published official Solutions Library sample as of 2026, so treat it as a vendor reference architecture rather than a battle-tested community standard, but it is the closest thing here to a real production blueprint: a seven-account layout for isolation and CI/CD promotion, native evaluation gates, and the five core AgentCore components, Runtime, Identity, Gateway, Memory, and Code Interpreter, wired together out of the box. A single-account tier exists for prototyping at roughly $5/month of light usage. Clone it when you are done learning the patterns and need to see how a cloud architect actually stages an agent fleet across dev, staging, and prod. Open AWS Guidance for Enterprise Agentic AI Platform on GitHub.
• AI Engineering Hub (patchy631). Ninety-three runnable projects graded beginner to advanced, from local OCR apps and first RAG workflows to DeepSeek fine-tuning, a from-scratch reasoning model, and a NotebookLM clone. At around 37k stars it has become the finish-a-project repo for the applied LLM stack, and its MCP shelf alone holds fifteen projects. We gave it a full Creator Pick walkthrough. Open AI Engineering Hub on GitHub.
• Awesome MCP Servers (punkpeye). The catalog of the MCP ecosystem: more than 3,300 server implementations across 58 categories, with badges marking official implementations, implementation language, and cloud or local scope. Check here before writing a tool integration by hand. We gave it a full Creator Pick walkthrough. Open Awesome MCP Servers on GitHub.

Guides: the two engineering essays that cut through the hype

There is no shortage of vendor blog posts about agents. These two are different: they come from the labs that built the models, they are written for engineers shipping real systems, and they are short enough to read in one sitting. Read them after you clone a repo and before you design your own loop.

• Claude Code Best Agentic Coding practices (Anthropic). The field guide for building coding agents, written by the team behind Claude Code, and the closest thing to a manual on how to prompt, scaffold, and scope an agent that writes software. It is unusually specific about what actually works in agentic coding, context management, tool selection, evaluation loops, which is why it earns a slot over the more generic framework docs. Read it once, then keep it open while you build. Read Claude Code Best Agentic Coding practices.
• OpenAI’s Practical Guide to Building Agents (OpenAI). A short, opinionated PDF that walks through when an agent is the right abstraction, how to scope a single-agent vs multi-agent setup, and what to guardrail before you ship. It is the complement to the Anthropic guide: where Anthropic goes deep on coding agents, OpenAI gives you the framing for deciding whether to build an agent at all. Read it as the decision layer that sits above the repos. Read OpenAI’s Practical Guide to Building Agents.

Videos: the talks that show an agent being built on screen

Some patterns are faster to learn by watching someone build them than by reading the code. These eight videos cover the range from LLM fundamentals to multi-agent systems. Pick the one closest to what you are stuck on, then come back for the rest.

• LLM Introduction (YouTube). A clear, lecture-style on-ramp to large language models for viewers who want the concepts before they see any code. Watch it first if you are coming from a non-ML background and need the mental model that the rest of this list assumes you have. Watch LLM Introduction on YouTube.
• LLMs from Scratch (YouTube). Builds a small language model step by step, which demystifies the box every agent sits on top of, and pairs well with Sebastian Raschka’s book of the same name. It is the video to reach for when you want to stop treating the LLM as a black box. Watch LLMs from Scratch on YouTube.
• Agentic AI Overview (Stanford) (Stanford Online). A grounded, academic overview of where the agentic AI field actually is, which is a useful corrective to the hype cycle most agent content lives inside. Watch it to reset your expectations about what agents can and cannot reliably do today before you commit to an architecture. Watch the Stanford Agentic AI Overview.
• Building and Evaluating Agents (AI Engineer). Gets into the part everyone skips: how you actually measure whether an agent is getting better or worse as you change its prompt and tools. It is the video to watch once you have a working prototype and need to stop guessing. The evaluation framing alone is worth the watch. Watch Building and Evaluating Agents.
• Building Effective Agents (AI Engineer). A talk-length treatment of the patterns that make agents work in production, echoing the Anthropic engineering essay but with the speaker building live on screen. Use it as the bridge from the repos above to the design decisions in your own system. Watch Building Effective Agents.
• Building Agents with MCP (AI Engineer). Walks through wiring an agent to tools via the Model Context Protocol, which is fast becoming the standard way to expose capabilities to a model. If your agent needs to call a database, a file system, or an external API, this is the talk that shows the pattern end to end. Watch Building Agents with MCP.
• Building an Agent from Scratch (AI Engineer). Strips an agent down to its loop, no framework, no abstraction, just the reasoning-act-observe cycle in plain code. It is the single best video for understanding what a framework like LangGraph or CrewAI is actually doing for you under the hood. Watch it before you adopt any agent SDK. Watch Building an Agent from Scratch.
• Philo Agents (The Neural Maze). A playlist that works through agent design over multiple episodes, deeper and slower than a single talk, which suits learners who want a course-style path rather than one-off videos. It is the natural next step once you have watched the standalone talks above and want the threads pulled together. Watch the Philo Agents playlist.

If you only have time for a few of these AI agents repos and their companions, start with GenAI Agents for the patterns, read OpenAI’s Practical Guide to decide whether you need an agent at all, and watch Building an Agent from Scratch to see the loop in plain code. That trio is enough to go from reading about agents to building one. Everything else extends that foundation: the bigger curricula for fundamentals, the Anthropic guide for coding agents, and the evaluation and MCP talks for when your prototype has to grow up.