Agentic AI Roadmap

August 21, 2026 · View on GitHub

A production-grade, hands-on curriculum for building agentic AI systems with LangGraph, MCP, and A2A. From first principles to enterprise deployment.

Star this repo to follow updates. To contribute, read CONTRIBUTING.md and STYLING.md.

What this is

A structured learning path that takes you from LLM fundamentals to production multi-agent systems. Every chapter ships with runnable code, a golden dataset, an evaluator, and a production-readiness checklist. Every project is specced to portfolio depth. The goal is not to teach you what agentic AI is — it is to make you an engineer who can ship agentic systems that work.

The curriculum is built around the 2026 reference stack: LangGraph for orchestration, MCP for tool interop, A2A for agent interop, LangSmith for evaluation and observability, and LangGraph Platform for deployment. These are the choices that have won in production. We teach them deeply, and we explain why they won.

Learning paths

Pick the path that matches your background. Each path tells you what you already know, what you need to learn, and the shortest route through the curriculum.

From ML engineer — easiest transition; add agents and evals to your existing stack
From backend engineer — add AI orchestration to distributed systems experience
From data scientist — evaluation is your superpower; add engineering rigor
From frontend engineer — backend first, then AI, then full-stack agents
From data engineer — pipelines and persistence map directly; add orchestration
From zero — the complete beginner path, no assumptions

Curriculum

The nine modules below are the spine of the roadmap. Each is self-contained with a README that orients you, prerequisites, and a clear "what's next" pointer.

01 Foundations — what agentic AI is, why LangGraph, LLM fundamentals, prompt engineering, structured outputs
02 LangGraph core — graph state, nodes, edges, conditional routing, parallel and iterative workflows, functional API
03 Agents in practice — conversational agents, persistence, tool integration, human-in-the-loop, streaming, Studio
04 Tools and MCP — tool design, MCP from scratch, building MCP servers, consuming MCP from LangGraph, tool registries
05 Agentic patterns — ReAct, plan-and-execute, reflexion, supervisor, swarm, hierarchical, map-reduce, when to use which
06 Evals and observability — golden datasets, LLM-as-judge, trajectory evals, LangSmith tracing, cost and latency, CI regression suites
07 Multi-agent and A2A — multi-agent architectures, handoffs, A2A protocol, agent cards, cross-framework interop
08 Production — LangGraph Platform, Docker, Kubernetes, checkpointing, cost optimization, governance, continuous improvement
09 Emerging topics — agent OS, long-horizon memory, autonomous SDLC, agentic browsers, self-improving agents, safety

Field guide

The field guide is a separate section modeled on Alexey Grigorev's AI Engineering Field Guide. It covers the role, the skills, the job market, the interview process, and career trajectory for agentic AI engineers in 2026. Read it alongside the curriculum — the curriculum teaches you what to build, the field guide teaches you what to do with what you build.

Projects

Ten portfolio-grade projects, each specced to portfolio depth: problem, architecture, stack, eval rubric, datasets, stretch goals, and references to real job postings that ask for the skill. Build these to consolidate the curriculum into a demonstrable portfolio.

See projects/ for the full catalog.