Picture by Creator | Canva & ChatGPT
# Introduction
GitHub has grow to be the go-to platform for learners desirous to study new programming languages, ideas, and expertise. With the rising curiosity in agentic AI, the platform is more and more showcasing actual tasks that target “agentic workflows,” making it an excellent surroundings to study and construct.
One notable useful resource is microsoft/ai-agents-for-beginners, which incorporates a 12-lesson course overlaying the basics of constructing AI brokers. Every lesson is designed to face by itself, permitting you to begin at any level that fits your wants. This repository additionally gives multi-language help, making certain broader accessibility for learners. Every lesson on this course consists of code examples, which could be discovered within the code_samples folder.
Furthermore, this course makes use of Azure AI Foundry and GitHub Mannequin Catalogs for interacting with language fashions. It additionally incorporates a number of AI agent frameworks and providers like Azure AI Agent Service, Semantic Kernel, and AutoGen.
To facilitate your decision-making course of and supply a transparent overview of what you’ll study, we’ll evaluate every lesson intimately. This information serves as a useful useful resource for learners who may really feel unsure about selecting a place to begin.
# 1. Intro to AI Brokers and Agent Use Instances
This lesson introduces AI brokers — methods powered by massive language fashions (LLMs) that sense their surroundings, motive over instruments and data, and act — and surveys key agent sorts (easy/model-based reflex, purpose/utility-based, studying, hierarchical, and multi-agent methods (MAS)) via travel-booking examples.
You’ll study when to use brokers to open-ended, multi-step, and improvable duties, and the foundational constructing blocks of agentic options: defining instruments, actions, and behaviors.
# 2. Exploring AI Agentic Frameworks
This lesson explores AI agent frameworks with pre-built parts and abstractions that allow you to prototype, iterate, and deploy brokers sooner by standardizing frequent challenges and boosting scalability and developer effectivity.
You’ll examine Microsoft AutoGen, Semantic Kernel, and the managed Azure AI Agent Service, and study when to combine together with your present Azure ecosystem versus utilizing standalone instruments.
# 3. Understanding AI Agentic Design Patterns
This lesson introduces AI agentic design ideas, a human-centric person expertise (UX) method for constructing customer-focused agent experiences amid the inherent ambiguity of generative AI.
You’ll study what the ideas are, sensible pointers for making use of them, and examples of their use, with an emphasis on brokers that broaden and scale human capacities, fill data gaps, facilitate collaboration, and assist folks grow to be higher variations of themselves via supportive, goal-aligned interactions.
# 4. Instrument Use Design Sample
This lesson introduces the tool-use design sample, which permits LLM-powered brokers to have managed entry to exterior instruments similar to capabilities and APIs, enabling them to take actions past simply producing textual content.
You’ll study key use circumstances, together with dynamic information retrieval, code execution, workflow automation, buyer help integrations, and content material era/enhancing. Moreover, the lesson will cowl the important constructing blocks of this design sample, similar to well-defined device schemas, routing and choice logic, execution sandboxing, reminiscence and observations, and error dealing with (together with timeout and retry mechanisms).
# 5. Agentic RAG
This lesson explains agentic retrieval-augmented era (RAG), a multi-step retrieval-and-reasoning method pushed by massive language fashions (LLMs). On this method, the mannequin plans actions, alternates between device/perform calls and structured outputs, evaluates outcomes, refines queries, and repeats the method till reaching a passable reply. It usually makes use of a maker-checker loop to boost correctness and recuperate from malformed queries.
You’ll study concerning the conditions the place agentic RAG excels, notably in correctness-first situations and prolonged tool-integrated workflows, similar to API calls. Moreover, you’ll uncover how taking possession of the reasoning course of and utilizing iterative loops can improve reliability and outcomes.
# 6. Constructing Reliable AI Brokers
This lesson teaches you learn how to construct reliable AI brokers by designing a sturdy system message framework (meta prompts, fundamental prompts, and iterative refinement), implementing safety and privateness greatest practices, and delivering a high quality person expertise.
You’ll study to determine and mitigate dangers, similar to immediate/purpose injection, unauthorized system entry, service overloading, knowledge-base poisoning, and cascading errors.
# 7. Planning Design Sample
This lesson focuses on planning design for AI brokers. Begin by defining a transparent general purpose and establishing success standards. Then, break down complicated duties into ordered and manageable subtasks.
Use structured output codecs to make sure dependable, machine-readable responses, and implement event-driven orchestration to handle dynamic duties and surprising inputs. Equip brokers with the suitable instruments and pointers for when and learn how to use them.
Repeatedly consider the outcomes of the subtasks, measure efficiency, and iterate to enhance the ultimate outcomes.
# 8. Multi-Agent Design Sample
This lesson explains the multi-agent design sample, which entails coordinating a number of specialised brokers to collaborate towards a shared purpose. This method is especially efficient for complicated, cross-domain, or parallelizable duties that profit from the division of labor and coordinated handoffs.
On this lesson, you’ll study concerning the core constructing blocks of this design sample: an orchestrator/controller, role-defined brokers, shared reminiscence/state, communication protocols, and routing/hand-off methods, together with sequential, concurrent, and group chat patterns.
# 9. Metacognition Design Sample
This lesson introduces metacognition, which could be understood as “interested by pondering,” for AI brokers. Metacognition permits these brokers to watch their very own reasoning processes, clarify their choices, and adapt primarily based on suggestions and previous experiences.
You’ll study planning and analysis strategies, similar to reflection, critique, and maker-checker patterns. These strategies promote self-correction, assist determine errors, and forestall countless reasoning loops. Moreover, these strategies will improve transparency, enhance the standard of reasoning, and help higher adaptation and notion.
# 10. AI Brokers in Manufacturing
This lesson demonstrates learn how to rework “black field” brokers into “glass field” methods by implementing sturdy observability and analysis strategies. You’ll mannequin runs as traces (representing end-to-end duties) and spans (petitions for particular steps involving language fashions or instruments) utilizing platforms like Langfuse and Azure AI Foundry. This method will allow you to carry out debugging and root-cause evaluation, handle latency and prices, and conduct belief, security, and compliance audits.
You’ll study what facets to judge, similar to output high quality, security, tool-call success, latency, and prices, and apply methods to boost efficiency and effectiveness.
# 11. Utilizing Agentic Protocols
This lesson introduces agentic protocols that standardize the methods AI brokers join and collaborate. We are going to discover three key protocols:
Mannequin Context Protocol (MCP), which offers constant, client-server entry to instruments, assets, and prompts, functioning as a “common adapter” for context and capabilities.
Agent-to-Agent Protocol (A2A), which ensures safe, interoperable communication and process delegation between brokers, complementing the MCP.
Pure Language Internet Protocol (NLWeb), which permits natural-language interfaces for web sites, permitting brokers to find and work together with internet content material.
On this lesson, you’ll study concerning the objective and advantages of every protocol, how they allow massive language fashions (LLMs) to speak with instruments and different brokers, and the place every matches into bigger architectures.
# 12. Context Engineering for AI Brokers
This lesson introduces context engineering, which is the disciplined apply of offering brokers with the appropriate data, in the appropriate format, and on the proper time. This method permits them to plan their subsequent steps successfully, transferring past one-time immediate writing.
You’ll find out how context engineering differs from immediate engineering, because it entails ongoing, dynamic curation relatively than static directions. Moreover, you’ll perceive why methods similar to writing, deciding on, compressing, and isolating data are important for reliability, particularly given the constraints of constrained context home windows.
# Remaining Ideas
This GitHub course offers the whole lot it is advisable begin constructing AI brokers. It consists of complete classes, quick movies, and runnable Python code. You’ll be able to discover subjects in any order and run samples utilizing GitHub Fashions (obtainable without cost) or Azure AI Foundry.
Moreover, you should have the chance to work with Microsoft’s Azure AI Agent Service, Semantic Kernel, and AutoGen. This course is community-driven and open supply; contributions are welcome, points are inspired, and it’s licensed so that you can fork and prolong.
Abid Ali Awan (@1abidaliawan) is an authorized information scientist skilled who loves constructing machine studying fashions. At the moment, he’s specializing in content material creation and writing technical blogs on machine studying and information science applied sciences. Abid holds a Grasp’s diploma in expertise administration and a bachelor’s diploma in telecommunication engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students fighting psychological sickness.
