What this is
The AI Path on StackNova has grown to cover the modern AI stack for builders: what a language model actually is and why it invents things, the model family you call and the platforms you call it through, the protocol and the retrieval pattern that connect it to your data, and the developer tools that put all of it in your daily workflow. This page is the map across that library, with a short note on each article - except the agent layer, which has a map of its own.
It is not a tutorial and not an opinionated comparison. It is the page you bookmark, send to a teammate who is catching up, or come back to when you are not sure which article to read next. For the agent layers specifically there is now a dedicated deep map - the Agentic AI stack - which this page links into rather than duplicates.
Pick a layer, read its article, follow the cross-links. The map is the index, the articles are the depth.- how to read this page
Foundations
Before any product: what the engine is, why it invents things, the vocabulary around it, and what its answers are doing to the way people find things.
Start with what a model is at all - the learned part of the system - then the umbrella term, what generative AI is and how composing differs from looking up, then the engine: what a large language model is and why it is stateless. Then the failure that follows from how it works - why a fluent answer can be false, and what a product does about it. The search piece closes the loop on what LLM-generated answers change about finding information at all.
- → What is a model? LEVEL 101 - the learned part of the system, and the vocabulary around it
- → What is Gen AI? LEVEL 101 - the umbrella term
- → What is an LLM? LEVEL 101
- → Why do AI models hallucinate? LEVEL 101
- → What is Augmentation? LEVEL 101 - four senses of one word
- → AI answers and search LEVEL 101
Models and platforms
The model layer you actually call, the catalog for going deep on it, and the platforms and plans you call it through - the app for people and the API for software are different products.
Claude is the model family on which most of the StackNova AI surface is built. The article covers the three-tier family (small / mid / large), what each tier is good for, and how to think about pricing, context windows, and the practical differences from other providers. The two learning pieces map the official free course catalog and the Academy that hosts it.
- → What is Claude? LEVEL 101
- → Learn Claude LEVEL 101
- → What is Claude Academy? LEVEL 101
Both vendors sell the same two things - an app for people and an API for software - in separate accounts with separate bills. Two articles cover the platforms you build on; three comparisons settle the recurring purchase questions: team plan or API, business chat or platform, and the two vendors side by side.
- → What is the Claude Platform? LEVEL 201
- → What is the OpenAI Platform? LEVEL 201
- → Claude Team vs Claude Platform LEVEL 101 - team plan or API
- → ChatGPT Business vs OpenAI Platform LEVEL 101 - staff app or software API
- → Claude vs OpenAI: plans and platforms LEVEL 201 - the two vendors, side by side
Protocols and retrieval
Two ways to get your data in front of the model: a protocol that lets the model call your tools itself, or a retrieval step your code runs before every call. Both live here, with the comparison that decides between them.
Model Context Protocol (MCP) is the open standard for connecting AI assistants to external tools, data sources, and services. It is the JSON-RPC-based contract Claude Desktop, Claude Code, Copilot, and a growing list of clients all speak. Understanding MCP is what lets you build once and integrate everywhere.
- → What is Model Context Protocol? LEVEL 101
- → Inside MCP LEVEL 301
Retrieval-augmented generation looks the answer up before the model writes it: the 800-page problem, the three-step fix, and what it costs, kept short; the 301 opens the pipeline part by part. The comparison puts it next to MCP - a pattern and a protocol, not rivals - and shows that the real choice is who triggers the search.
- → What is RAG? LEVEL 101
- → Inside a RAG pipeline LEVEL 301 - chunking, embeddings, BM25, fusion
- → RAG vs MCP: pipeline or tool? LEVEL 201 - who triggers retrieval
Agents: the other map
The agent layer has a map of its own. This page hands off rather than repeats.
An agent is a model running in a loop with tools, memory, and a goal - the layer where teams now spend most of their engineering time, and the one this page deliberately does not list. The idea, the context window, the extension mechanisms, the architectures past one agent, evaluation, failure modes, cost, and the frameworks all live in one dedicated map, in the order the ideas depend on each other.
- → Agentic AI stack - six layers, three reading orders
Developer tools
The surface most builders interact with daily: editor assistants, terminal agents, a delegated-work app, design helpers, and image generators. Tools on top of the models, packaged for specific jobs.
Two AI pair programmers, two different shapes. Claude Code is an agentic session that takes a goal end-to-end; GitHub Copilot lives inside your editor with completions, chat, and agent mode. The three companion pieces compare them - the side-by-side, the temperament essay, and a real migration told screenshot by screenshot. The CLAUDE.md piece covers the one file Claude Code reads at the start of every session.
- → What is Claude Code? LEVEL 201
- → What is a CLAUDE.md file? LEVEL 101
- → What is GitHub Copilot? LEVEL 101
- → GitHub Copilot and Claude Code LEVEL 101 - the side-by-side
- → Copilot vs Claude Code: engineer vs artist LEVEL 101 - the temperament essay
- → Claude Code is a Magician LEVEL 101 - a real migration, twenty minutes
Claude Cowork is the delegated-work surface built on the same agentic architecture as Claude Code, with no terminal in sight - you hand it files and a goal. The comparison piece settles which of the two agentic surfaces fits which job.
- → What is Claude Cowork? LEVEL 101
- → Claude Code vs Claude Cowork LEVEL 201
Claude Design is Anthropic's web surface for using Claude as a design partner. For images, two model families from two vendors: Nano Banana (Gemini 2.5 Flash Image), Google's fast, conversational image-generation and editing model, and GPT Image, the models behind image creation in ChatGPT, exposed to developers through the OpenAI Platform.
- → What is Claude Design? LEVEL 101
- → What is Nano Banana? LEVEL 101
- → What Nano Banana can do LEVEL 101 - the practical capability tour
- → What is GPT Image? LEVEL 101
Four reading orders
The map is sorted by layer, but most readers want a route. Four orders, depending on who you are.
The shortest way in
- What is a model? → What is Gen AI? → What is an LLM? + why it hallucinates
- What is Claude?
- What is GitHub Copilot?
- GitHub Copilot and Claude Code
- Try Copilot inside your editor for a week
Broaden the toolkit
- What is Claude Code? + CLAUDE.md
- Copilot vs Claude Code: engineer vs artist
- Claude Code is a Magician
- What is Claude Cowork? for the work around the code
- Claude Design + Nano Banana
- MCP to wire your own tools in
You want to build the agent, not just use one
- What is an LLM? + why it hallucinates
- What is Model Context Protocol?
- What is RAG? → RAG vs MCP
- The Claude Platform or the OpenAI Platform
- Then the Agentic AI stack map, start to finish
You want the whole stack, in order
- The vocabulary + the engine
- What is Claude? (the model)
- MCP (the wire) + RAG (the lookup)
- Then the Agentic AI stack map (the loop)
- Then the AI Path index, newest first
The map at a glance
If you read only this section, here is the AI stack from the bottom up:
- Foundation: what a model is, what generative AI is, what an LLM is, why it hallucinates, and what AI answers do to search.
- Model: Claude (and other foundation models you call via API), plus the course catalog.
- Platform: the Claude Platform and the OpenAI Platform - the API side, sold apart from the apps; plans and platforms compared.
- Protocol and retrieval: MCP for tool and data integration, RAG when your code does the looking up - RAG vs MCP decides.
- Agents: the loop, its window, its extensions, the architectures, and the operations - all in the Agentic AI stack map.
- Tools: GitHub Copilot in the editor, Claude Code in the terminal, Claude Cowork for delegated knowledge work, Claude Design for design, Nano Banana or GPT Image for images.