Learning AI & Machine Learning

AI (Artificial Intelligence) is a field no engineer can ignore in 2026. With generative AI tools like ChatGPT, Claude, and Gemini now part of everyday workflows, understanding what AI is doing and how it works has become a real advantage for engineers.

This section covers everything from AI and machine learning fundamentals to generative AI, RAG, the AI agent stack — memory, tools, MCP, skills, hooks, subagents, and orchestration — fine-tuning, and evaluation, all in a way that builds knowledge step by step.

What I Cover Here

AI Transformation

Go beyond AI and ML technical understanding to learn AI adoption at the organizational and business level. Structured across five phases: strategy, readiness, organization building, execution, and sustainment.

AI Transformation — Section entry point covering the definition, DX differences, and learning order
Organizational and Culture Transformation in AI — Explains organizational design and culture transformation for successful AI transformation, including CoE, federated, and embedded models and a staged capability-building path.
What is AI COE — Role, structure, and how to launch a Center of Excellence for AI
AI-Driven vs. AI-Native Organizations — Defines and contrasts the two organizational models — AI-Driven and AI-Native — from the perspectives of strategy, culture, technology, and operations.
AI Driven vs. AI Native Development — Compares the development approaches of AI Driven and AI Native organizations across development processes, technology stacks, team structures, and decision-making styles. Provides a framework for choosing which development model to pursue.
Individual AI Use vs. Organizational AI Use — Individual and organizational AI adoption differ fundamentally in purpose, strategy, and approach. Learn the characteristics of each and how to convert individual success into organizational value.
Talent and Skills Transformation in the AI Era — Explains the skill sets and talent strategy needed in the AI era, including how to combine internal development, hiring, external support, automation, reskilling, and new AI-era roles.
Individual AI Use Level Definitions — Defines five levels of individual AI use: Introduction, Validation, Application, Efficiency, and Transformation, from trying AI chat to improving work and spreading effective practices.
AI Maturity Model: Diagnosing Your Organization’s AI Evolution — A five-stage diagnostic synthesis informed by Gartner, IBM, and NIST
What is AI Ready? — Understanding AI Ready: the prerequisites for organizations to successfully adopt AI. Evaluate and improve your readiness across tools and technology, people, pilot projects, governance, diffusion, data, and infrastructure.
What is AI Powered? — Characteristics of an AI Powered organization: embedding AI into core business processes, measuring AI impact, and designing human-AI collaboration models.
AI Adoption, AI Enablement, and AI Transformation: Differences and Relationships — Explains the definitions and dependencies of AI adoption, AI enablement, and AI transformation, and shows how strategy, organization, and investment change depending on the level an organization aims to achieve.
What Is AI Native? — Explains AI Native as the end state in which AI becomes the default rather than the exception. Places it within the AI Ready to AI Transformation to AI Native progression, and covers the definition, five characteristics, diagnostic questions, and paths for established companies.
What Is Agent Ready? — Explains Agent Ready as the state in which AI agents can act autonomously within business operations. Covers how it differs from AI Ready, the six requirements of connectivity, identity, codified procedures, context supply, observability, and guardrails, plus a self-assessment checklist.
AI Transformation Strategy Frameworks — Covers how to build an AI transformation strategy using McKinsey’s Rewired framework, BCG’s AI transformation portfolio, and Accenture’s Total Enterprise Reinvention — including how to prioritize initiatives.
Change Management for AI Transformation — Covers the theory and practice of change management for successful AI transformation — applying Kotter’s 8-step model, resistance patterns and responses, communication strategy, and Microsoft’s growth mindset case study.
The AI-Driven Operating Model: An Overview — Explains what an AI-first operating model is, how it differs from traditional models, and the three core elements of transformation — drawing on McKinsey’s Agile Operating Model and Deloitte’s Human-Machine Collaboration Model.
AI Adoption Roadmap — Explains a three-phase roadmap and practical methodology for moving beyond ad-hoc PoCs and turning AI adoption into real business value.
Preventing PoC Failure — Production-Backward design, countering the five root causes, and production rollout conditions
Sustaining AI: Communities and Organizational Self-Directed Learning — Explains how to design study groups, communities of practice, and self-directed learning systems that help organizations sustain AI adoption.

AI & Machine Learning Basics

Learn how AI, machine learning, and deep learning relate to each other, and understand the core concepts behind each.

What Is Machine Learning? — How machines learn from data, and how this differs from traditional programming
What Is Deep Learning? — How neural networks work, and an overview of LLMs (Large Language Models)
Learning Paradigms — Transfer learning, fine-tuning, federated learning, and other modern techniques

How Generative AI Works

Learn the foundations of generative AI itself: LLMs, reasoning models, and how text, image, music, and video generation work.

How Generative AI Works — The big picture of AI that generates text, images, music, and video
What Is an LLM? Large Language Models Explained — How large language models work and developed
Transformer Models — Starting from the 2017 paper ‘Attention Is All You Need,’ this article explains the Transformer’s Self-Attention, Multi-Head Attention, and Positional Encoding for beginners. Learn the mechanisms that underpin modern LLMs, with diagrams.
BERT vs. GPT — Both BERT and GPT are Transformer-based, but their architectures are opposites. This article explains the differences between bidirectional vs. unidirectional, MLM vs. autoregressive — and which tasks each excels at — for beginners.
Reasoning Models — Explains how ‘think before answering’ reasoning models and modes work. Covers Chain-of-Thought, RLHF, and related techniques, and how they differ from standard LLMs — for intermediate learners.
Generative AI Models and Intelligence Metrics — Model types, IQ-style scores, and practical capability signals
How Text Generation Works — A beginner-friendly explanation of how LLMs generate text. Covers token prediction, sampling strategies, context windows, and prompt design — with ChatGPT, Claude, and Gemini as concrete examples.
How Image Generation Works — A beginner-friendly explanation of how AI image generation works, focused on diffusion models. Uses DALL-E, Midjourney, Stable Diffusion, and Imagen as examples to explain how Text-to-Image AI works.
How Video Generation Works — A beginner-friendly explanation of AI video generation. Covers Video Diffusion Models, the DiT architecture, and how temporal consistency is achieved.
How Music Generation Works — A beginner-friendly explanation of AI music generation. Covers token-based music generation, diffusion model approaches, and how AI learns musical structure.

Generative AI Engineering

Learn how to design AI instructions, context, execution environments, and repeat structure for practical use: prompt engineering, context engineering, harness engineering, and loop engineering.

Prompt Engineering — Design instructions that make answer quality more stable
Context Engineering — Provide the documents, history, and constraints AI needs
Harness Engineering — Connecting AI to tools, permissions, and checks
Loop Engineering — Designing an iteration’s granularity, exit conditions, and pacing
The AI Agent Stack — An overview of the twelve representative layers that make up a modern AI agent — RAG, Context, Memory, Agent, Agentic AI, Tools, MCP, Skills, Hooks, Subagents, Orchestration, and Eval — with links to deeper explanations of each.
The Context Window and Its Limits — Explains how the context window works: what counts toward the limit, why filling it degrades quality, and what to do as the limit approaches.
What Is Prompt Caching? — Explains how shared-prefix computation is reused, with distinctions among Claude, the OpenAI API, ChatGPT, and Codex.
What Is Context Compaction? — Explains how a long conversation is reduced to needed task state, using Claude Code, Codex, and ChatGPT examples and covering information-loss risks.
Why Quality Degrades in Long Conversations — Explains why AI quality degrades in long conversations — instruction dilution, stale information, and accumulated contradictions — and how to counter each.

RAG (Retrieval-Augmented Generation)

Learn how RAG lets LLMs search external documents and answer from evidence. This is the foundation for safely connecting AI to internal documents, current information, and specialized sources.

What Is RAG? — RAG flow, architecture, evaluation, and current best practices
The History of RAG — The path from search and QA systems to modern RAG
RAG Architecture Patterns — Naive RAG, Advanced RAG, Graph RAG, and when to use each
What is Agentic RAG? — Search planning, re-search, and verification handled by agents
Code RAG and Coding Agents — Repository search, structured chunks, and test execution
The Future of RAG — How RAG will evolve, examined through the lens of long-context models, Agentic RAG, Code RAG, Graph RAG, permission design, and evaluation.
Embeddings & Vector Representations — How text becomes vectors for semantic search
Retrieval Strategies — BM25, vector search, hybrid search, and reranking
Chunking Strategies — How to split documents into retrievable units
Choosing a Vector Database — Comparing vector stores for RAG systems

Memory

Learn how an AI agent retains information across conversation turns and sessions, and recalls it when needed.

What Is Memory — How it differs from context, and the roles of short-term and long-term memory
Short-Term and Long-Term Memory — Lifespan and write criteria for each type
Memory Versus RAG — How to choose between two similar mechanisms
Long-Term Memory Storage Patterns — Comparing files, structured records, vector search, and summaries

AI Agents

Learn how AI systems use tools, plan work, and complete multi-step tasks.

What Is an AI Agent? — Core components and the ReAct loop
AI Agents and MCP — How MCP connects agents to external tools

Agentic AI

Learn the idea of agentic AI: systems that plan, decide, and self-correct while working toward a goal.

What Is Agentic AI — How it differs from “agent,” and the three elements of agentic behavior
Planning and Task Decomposition — Breaking a goal into executable steps
Self-Correction and Reflection — Checking results and adjusting the approach
Designing Autonomy Levels — Staging how much to hand over to AI

Tools

Learn how AI agents call functions to connect with the outside world and actually act.

What Is Tool Use — Elements of a tool definition and the principle of least privilege
Tool Selection Failures and Fixes — Why an AI picks the wrong tool, and how to fix it
Execution Permissions and Approval — Least-privilege design and where approval belongs
Tool Error Handling — Error message design and retry policy

MCP (Model Context Protocol)

Learn about MCP — the open standard that lets AI models connect with external tools and data. Understanding MCP explains why tools like Claude Code and Claude Desktop can integrate with so many services.

What Is MCP? — An overview of the protocol that bridges AI and external tools
Why MCP? — The M×N integration problem and how MCP solves it
MCP Architecture — The three-layer structure: Host, Client, and Server
MCP Capabilities — Tools, Resources, and Prompts — what they are and when to use each
Remote MCP vs Local MCP — An explanation of the two types of MCP server deployment — Local MCP and Remote MCP — using Figma’s Native MCP as a concrete example. Covers how each works, their differences, and real-world Skills (capabilities) for beginners.

Skills

Learn the idea of skills: reusable know-how loaded on demand, rather than kept in context at all times.

What Are Skills — How skills differ in role from prompts and tools
Loading Only When Needed — Progressive disclosure and how to write the summary
Skills Versus Prompts Versus MCP — How the four mechanisms differ in role
Writing a Good Skill — Guidance on granularity, steps, and completion conditions

Hooks

Learn how hooks run logic automatically at predetermined points in an agent’s execution.

What Are Hooks — Using hooks for verification, permission control, and logging
Hook Trigger Points — What each timing makes possible
Hook Use Cases — Quality gates, safety checks, recording, consistent baseline
Points to Note When Using Hooks — Infinite loops, latency, over-blocking

Subagents

Learn how a main agent delegates specific work to specialized helper agents.

What Are Subagents — How context isolation improves quality
Delegation Patterns — Choosing between research, review, and parallel patterns
Context Isolation — Why isolation improves quality, and its caveats
When Not to Use Subagents — Cases where delegation becomes inefficient

Orchestration

Learn how to coordinate multi-step workflows and multiple AI agents so they add up to one completed job.

What Is Orchestration — Splitting work, assigning handlers, controlling flow, verification and recovery
AI Agent Orchestration — Comparison of single-agent, multi-agent, and hierarchical structures
Comparing Orchestration in Claude, ChatGPT, and Codex — Controls and request examples across each service
Parent-Child Agent Design in Claude Code — A capable parent, task-specific subagents, Skills, and worktree examples
Parent-Child Agent Design in Codex — A high-effort parent, lightweight subagents, AGENTS.md, and Skill examples
AI Agent Framework Comparison — LangGraph, OpenAI Agents SDK, Claude Agent SDK, Google ADK, and other options
State and Handoff — Designing how information passes between steps
Failure Detection and Recovery — Detecting and recovering from mid-process failures

AI Evaluation

Learn how to measure model and LLM-application quality before production use.

What Is AI Evaluation? — Automated evaluation, human evaluation, and LLM-as-a-Judge
AI Evaluation Frameworks — LM Evaluation Harness, Ragas, DeepEval, and related tools
Business Fit Evaluation — Task completion, cost, and ROI in real workflows
Safety & Harm Evaluation — Harmful output, prompt injection, and leakage risk checks
Consistency & Reliability Evaluation — Testing whether outputs stay stable and dependable
Human-in-the-Loop Evaluation — Where human judgment belongs in evaluation pipelines
Evaluating AI Agents — Evaluating the execution trajectory, not just the final output
Tracing and Observability — Recording agent behavior so it can be traced after the fact

Fine-tuning

Learn how additional training changes an LLM’s behavior, style, and task performance when prompting or RAG is not enough.

What Is Fine-tuning? — Core concepts and when to choose fine-tuning versus RAG or prompts
Fine-tuning Methods Compared — Full Fine-tuning, LoRA, QLoRA, and Adapter layers
Training Data Preparation — Collecting, formatting, and checking training data
Knowledge Distillation — Transferring large-model behavior into smaller models

AI Governance

Learn the decision-making structures, policies, and processes for using AI safely, ethically, and legally. Covers the five principles of Responsible AI, major frameworks including NIST AI RMF, EU AI Act, and ISO/IEC 42001, and agile governance approaches for the fast-moving generative AI era.

What Is AI Governance? — Core concepts, major frameworks, and the three layers of governance
What Is Responsible AI? — The five core principles, types of AI bias, and an overview of EU AI Act and NIST AI RMF
Agile AI Governance — Practical approaches including Living Policy, sprint-based risk management, and continuous compliance
Human-in-the-Loop vs. Human-over-the-Loop — Explains the difference between Human-in-the-Loop and Human-over-the-Loop, how Human-on-the-Loop relates to them, and how to choose oversight patterns by AI risk level.
AI and Copyright: Comparing Legal Frameworks in Japan and the US — A structured comparison of copyright law in Japan and the United States as it applies to AI-generated content. Covers training data legality, ownership of AI outputs, and practical steps for corporate compliance.
Generative AI and Personal Information: Legal Obligations and Practical Compliance — A structured guide to legal risks when generative AI handles personal information. Covers Japan’s Act on the Protection of Personal Information (APPI) and the GDPR across three key scenarios: training data, prompt input, and AI output management.
Generative AI and Privacy: Risks and Design Principles — A systematic look at the privacy risks inherent to generative AI systems. Covers memorization, inference attacks, and contextual integrity, then walks through Privacy by Design principles and technical privacy-preserving techniques.
Rights Issues with AI-Generated Images: Copyright, Portrait Rights, Trademarks, Design Rights, Similarity, and Provenance — Understand the copyright, portrait rights, trademark, and design rights risks in AI-generated images, and learn practical enterprise responses based on similarity and provenance analysis.
Copyright and Risks of AI-Generated Code: OSS Licenses, Copyleft, and Enterprise Best Practices — Understand copyright ownership, OSS license risks, and copyleft concerns around AI-generated code, and learn what enterprises should do to safely leverage generative AI coding tools.
Balancing Guardrails and Governance for Enterprise Generative AI — A practical guide to balancing guardrails (technical and operational controls) with governance (organizational decision-making and policy) to promote safe generative AI adoption while managing risks.

Generative AI Security

Understand the security risks specific to generative AI and learn systematically from attack techniques through defense frameworks and guardrail implementation. Covers the knowledge developers need when embedding AI into products.

Key Attack Techniques — How prompt injection, jailbreaking, data poisoning, and model inversion attacks work
Security Frameworks — Comparing OWASP LLM Top 10, NIST AI 600-1, and MITRE ATLAS
OWASP Agentic AI Framework — 10 threat categories and mitigations specific to AI agents
Agent Security — Tool misuse, context poisoning, and MCP security design
How Guardrails Work and How to Implement Them — Input, output, and execution guard design patterns with implementation examples
IAM Authentication in the Agent Era — Explains IAM authentication in the agent era through human, workload, and AI agent identities, role-based authorization, delegation, audit logs, and enterprise examples.

Suggested Learning Order

Starting with AI & machine learning basics

Begin with What Is Machine Learning? to get a clear picture of how AI, ML, and deep learning relate before diving into each topic.

What Is Machine Learning? — AI vs. ML, and the three types of learning
What Is Deep Learning? — Neural networks and how LLMs work
Learning Paradigms — Practical techniques like transfer learning and fine-tuning

Using generative AI in practical work

Start with how LLMs work, then move through model choice, prompts, context, harnesses, RAG, and the AI agent stack. This path connects chat-based use to practical system design.

How Generative AI Works — The overall picture
What Is an LLM? — LLM basics
Generative AI Models and Intelligence Metrics — How to choose models
Prompt Engineering — Instruction design
Context Engineering — Supplying needed information
Harness Engineering — Execution and verification design
Loop Engineering — Designing the repeat structure
What Is RAG? — Evidence-grounded generation from external documents
What Is Memory — Retaining information across sessions
What Is an AI Agent? — AI systems that act across multiple steps
What Is Agentic AI — The plan, act, and self-correct loop
What Is Tool Use — Calling external functions safely
What Is MCP? — The standard for connecting tools and data
What Are Skills — Designing reusable procedures
What Are Hooks — Building in automatic verification and approval
What Is Orchestration — Coordinating multi-step, multi-agent work
What Is AI Evaluation? — Measuring quality before production use

Learning MCP for practical AI tool use

If you want to get more out of tools like Claude Code, start with What Is MCP?.

What Is MCP? — Core concepts and overview
Why MCP? — The problem it solves
MCP Architecture — How it works under the hood
MCP Capabilities — What you can actually do with it

Prerequisites

No advanced math is required. Having a basic sense of functions and probability from high school math will help, but it is not necessary to follow along with the conceptual explanations.

Familiarity with engineering basics (terminal, Git) is helpful but not required. Reading Engineering Basics first will give you a more practical foundation.