AI Series Index
Three tracks by difficulty and audience. Later articles reference earlier ones.
Fundamentals
Build your AI mental model from zero — structured from easy to hard
Reading Guide: Beginner Path vs Professional Path
First time here? This article tells you what to read, how to read it, and what you can skip—pick a path based on your background.
What Is AI? Three Metaphors That Make It Click
No formulas, no jargon pile-up—three everyday metaphors to build a first-principles understanding of artificial intelligence.
How LLMs "Talk": Tokens, Context, and Probability
Three core mechanisms—tokenization, context windows, and probabilistic sampling. Understand these and AI stops feeling like magic.
Prompt Basics: How to Ask AI and Get Good Answers
Five practical techniques—from casual questions to reliable output. No spellbooks—just clarity about what you want.
What Is RAG? Let AI Read Your Docs Before Answering
Retrieval-Augmented Generation end to end: chunking, embedding, retrieval, generation—one diagram and clear steps.
AI Agents: From Chat to Getting Work Done
Agents plan steps, call tools, and execute autonomously. Understand the loop and you'll see what Cursor and Devin are doing.
Fine-Tuning vs RAG vs Long Context: How to Choose
Three ways to make AI more domain-aware—when each fits, what it costs, and where it breaks. One decision table.
Multimodal AI: Eyes and Ears for Models
Text, images, audio, video—how multimodal AI unifies them, what works today, and what still doesn't.
AI Production Checklist: From Demo to Product
Ten checks for shipping AI in real products—security, cost, eval, monitoring. None optional.
Model Landscape
Text, vision, audio, multimodal, robotics — and embeddings / vector DBs
Text Models Landscape: From GPT to the Open-Source Wave
LLM capabilities, mainstream model comparison, closed vs open source—the 2026 text model map in one article.
Vision Models: From Understanding Images to Generating Worlds
CLIP, GPT-4V, Stable Diffusion, Sora—two vision AI paths (understanding vs generation) and engineering use cases.
Music and Audio Models: From Whisper to Suno
Speech recognition, synthesis, and music generation—three audio AI product lines with tech basics and real deployments.
Native Multimodal: Unified Model Architecture Today and Tomorrow
From pipeline glue to any-to-any—GPT-4o, Gemini, and Chameleon dissected at the architecture level.
Robotics and AI: From LLMs to VLA Models
Embodied AI, vision-language-action models, sim-to-real—how AI learns to act in the physical world in 2026.
Embeddings and Vector Databases: The Memory Layer for AI
How embeddings work, how vector DBs store and search them, and where they power RAG, semantic search, and recommendations—with Pinecone, pgvector, and Chroma.
Frontier Tech
MCP, Skills, Harness, context engineering — industry-standard concepts
MCP Intro: The USB Port for AI and Tools
What Model Context Protocol is, why it matters, and how to use it—real configs for Claude Desktop and Cursor.
MCP Deep Dive: Protocol Architecture and Server Development
JSON-RPC flow, stdio vs SSE, Tools/Resources implementation—and a minimal MCP server in TypeScript.
Agent Skills: Reusable AI Capability Modules
Cursor Skills and Claude Agent Skills—pack domain knowledge into modules agents load on demand. How they fit with MCP and system prompts.
Agent Harness: The Operating System for Agents
What a harness is, its components, and how Cursor, Claude Code, and Devin differ—an engineering deep dive.
Context Engineering: The Real Moat in the Agent Era
Beyond prompt engineering—how you select, organize, and compress context drives ~90% of agent performance.
Eval and Observability: Knowing When Agents Get Better
Offline eval, online monitoring, regression tests for agents—practical paths with LangSmith and Braintrust.
Case Studies
Architecture decisions in Cursor, Claude Code, Shopify, and OpenAI
Case Study: How Cursor Wires MCP + Skills + Harness Together
From Tab completion to Agent mode—product architecture and why Cursor is today's best AI coding harness.
Case Study: Claude Code's Long-Running Agent Design
How Anthropic's Claude Code manages 50+ step coding tasks—context compression, permissions, and CLAUDE.md project memory.
Case Study: Shopify's RAG Path to AI Support
How Shopify built Sidekick with RAG + agents—from MVP to ~82% automation, with real engineering milestones.
Case Study: OpenAI Responses API and Agent Orchestration
How OpenAI's 2025 Responses API unifies chat, tools, and code interpreter—and how it compares to the MCP route.