Agentic AI has emerged as a main driver of AI application worldwide. The high degree of autonomy inherently built into agentic systems also poses specific challenges. On this site you find my best picks on AI agents, Multi-Agent-Systems, and full-fledged Agentic AI applications. The list covers concepts, protocols, core SDKs and Frameworks, platforms, and more.
You want to start approaching Agents and Agentic AI by understanding their characteristics, how they are defined and differentiated and what concepts they are based upon. This OECD publication does all of this and more in a concise and well-written manner.
A beginner-friendly guide explaining agent design patterns, when to use agents, tool use, workflows, and evaluation. While it is not the most up-to-date publication on the topic, I still find this to be one of the best introductions to building agents and it still holds up perfectly.
The de-facto standard open protocol for connecting AI applications to tools, data sources, and external systems. Strong adoption (e.g. Claude Desktop, Cursor, Windsurf, Replit). This site by LF Projects, LLC has everything you need to develop with MCP.
Framework for building agentic applications with tool use, handoffs, tracing, and structured workflows. Enables (almost) everyone to build multi-agent workflows. OpenAI’s site is comprehensive and covers all relevant aspects.
Open‑source Python framework for complex multi‑agent workflows with persistent state management, built as an extension of LangChain. Enables structured agent coordination, deterministic execution, and scalable production‑grade graph architectures
Langgraph is not the only helpful framework on the market. Here is another framework for role-based multi-agent collaboration, task delegation, processes, and enterprise automation workflows. They differ in some aspects – check out the many comparisons online for your pick.
Open‑source framework by Microsoft for building coordinated multi‑agent systems through structured dialogue, role‑based collaboration. Enables controlled task delegation, flexible orchestration, and reproducible workflows.
Open‑source agentic coding system designed for autonomous software engineering, combining LLM‑driven reasoning with real execution environments. Runs agents that complete entire coding tasks
Every single role involved in AI agents and Agentic AI should – in my view – have a basic understanding of the security risks. If you need a comprehensive, hands-on report on securing and governing autonomous AI systems, this is it. As simple as this.
Sure, this governance framework might not be of interest for everyone. But it is the world’s first attempt (at least from an official state authority) to proposa a way to actually govern Agentic AI and the current updates led to more refined and extended content. One of a kind.