The learning plan — AI agent specialist
Six phases, built from scratch. Microsoft · Google · NVIDIA · Anthropic · AWS · freeCodeCamp — direct from the big players. Certifications marked throughout.
✦ plan designed by Claude (Anthropic) · I'll build this myself one dayA brilliant free introduction to the world of AI agents — and the perfect starting point before a single line of code is written. Work through all 12 episodes with Obsidian open, taking notes in your own words. Skip every code section entirely — you don't have Python yet and that's completely fine. The goal here is building your mental map: what agents are, how they think, what problems they solve. You'll return to the code sections after Phase 2 once Python is under your belt, and they'll make complete sense at that point.
Covers agent design patterns, AutoGen, Semantic Kernel, RAG, multi-agent systems, and tool use. Watch everything, note everything, skip all code snippets. A second pass for the code comes later in Phase 4.
learn.microsoft.com/shows/ai-agents-for-beginners ↗Everything in this plan runs on Python. Get solid here before moving on — it pays back every hour you invest. The Anthropic GitHub notebooks layer in naturally from week 3 onwards. Once complete, return to the Microsoft Phase 1 code sections — they'll make perfect sense now.
15 real projects built in-browser. The most structured free Python cert available — project-based from day one. Your primary Python course.
freecodecamp.org/learn/python-v9 ↗Jupyter notebooks covering the Claude SDK, API keys, streaming, and multimodal prompts. Start these in week 3 alongside Python — they reinforce each other perfectly.
github.com/anthropics/courses ↗Return to the 12 Microsoft episodes now that Python makes sense. Work through the Semantic Kernel and AutoGen code sections you skipped in Phase 1 — a completely different experience with Python under your belt.
github.com/microsoft/ai-agents-for-beginners ↗Build the conceptual foundation across multiple big players before going deep. These courses are deliberately non-technical or lightly technical — they're about understanding the landscape.
Three-course path: Intro to Gen AI, Intro to LLMs, and Responsible AI. Free on Google Cloud Skills Boost — earns skill badges. Join the GEAR programme for 35 free monthly credits.
skills.google/paths/118 ↗Microsoft's open-source GitHub course covering how LLMs work, RAG, fine-tuning, and responsible AI. Python code examples throughout. The best free conceptual deep-dive available.
github.com/microsoft/generative-ai-for-beginners ↗NVIDIA's free, no-code intro to generative AI — how it works, use cases, and limitations. Short (~2 hrs) but earns an NVIDIA DLI certificate. A great first NVIDIA credential.
learn.nvidia.com ↗Now you know how LLMs work, learn to talk to them properly and start building with the Claude API. Anthropic takes centre stage here as a focused thread — not the whole plan.
Core features, capabilities, and limits. Do this first — sets the vocabulary for everything in the Academy that follows.
anthropic.skilljar.com ↗Anthropic's 4D framework for designing, evaluating and iterating on prompts. Foundational for everything that follows in the developer track.
anthropic.skilljar.com ↗9-chapter hands-on notebook course direct from Anthropic. Covers basic to advanced: chain-of-thought, tool use, RAG. The practical companion to AI Fluency above.
github.com/anthropics/prompt-eng-interactive-tutorial ↗8+ hrs. System prompts, tool use, context windows, RAG, architecture patterns. The most important single Anthropic course in this plan — this is how you build real AI products.
anthropic.skilljar.com ↗This is where you go from "I understand AI" to "I can build agents". Three major players, three complementary perspectives — Microsoft on frameworks, Google on cloud-native agents, NVIDIA on production RAG systems.
12 lessons with code: agent design patterns, tool use, RAG, multi-agent systems, AutoGen, Semantic Kernel, and Azure AI Foundry. Works with free GitHub Models.
microsoft.github.io/ai-agents-for-beginners ↗Microsoft's lab-assessed credential. Scenario-based assessment proving you can build agents in Copilot Studio. Free to earn — highly relevant for enterprise and freelance client work.
learn.microsoft.com/credentials ↗Multi-agent systems, Agent Development Kit (ADK), MCP integration, and enterprise deployment on Google Cloud. Hands-on labs in real GCP environments via GEAR programme credits.
skills.google/paths/3273 ↗NVIDIA's flagship agent course. Covers dialog management, document reasoning, state management, and deploying agentic systems at scale. Earns a DLI certificate on assessment completion.
courses.nvidia.com ↗Build MCP servers and clients in Python. Tools, resources, and prompts — the three primitives that let any AI agent talk to the outside world.
anthropic.skilljar.com ↗Where do your agents live? Lead with Azure — Microsoft's free learning paths are the strongest and AZ-900 is the most recognised entry cloud credential. Google Cloud and AWS bolt on after.
Free study path covering cloud concepts, Azure services, and pricing models. The most widely recognised entry-level cloud cert globally. Study is free; exam is ~£90.
learn.microsoft.com — AZ-900 ↗Covers AI workloads, ML concepts, and Azure AI services. Natural follow-on from AZ-900 and directly relevant to deploying agents. Study free, paid exam ~£90.
learn.microsoft.com — AI-900 ↗Deploy Claude via AWS Bedrock — RAG pipelines, tool use, batch processing, and production patterns in the AWS ecosystem. Directly from Anthropic, free cert included.
anthropic.skilljar.com ↗Deploy Claude via Google Vertex AI — the full Claude API on GCP. Pairs with the Agentic AI path from Phase 4 to give you a complete GCP agent picture.
anthropic.skilljar.com ↗Amazon's free foundational cloud course. AWS is dominant in enterprise client work — add this after Azure. The Cloud Practitioner cert is the other major entry-level cloud credential alongside AZ-900.
aws.amazon.com/training ↗