Modern AI is not one product. It is a stack of parts working together — like a restaurant kitchen where the chef, the pantry, the recipe book, and the security guard all have different jobs. A popular diagram of this “AI ecosystem” has been going around LinkedIn (credit to Chandan S Roniyan for the original mapping). Here is what each of the 10 parts actually does, in plain English — with the real tools used for each job.
01🧠 LLM — the brain
The Large Language Model is the part that reads, thinks, and writes. ChatGPT, Claude, and Gemini are the famous ones; Llama and Mistral are versions companies can run on their own computers. Everything else on this list exists to help this brain do its job.
02📋 Agentic AI — the planner
An LLM answers one question. Agentic AI frameworks let it plan several steps: “first search, then compare, then write a report.” Think of it as giving the brain a to-do list and permission to work through it.
AWS Strands03📖 RAG — the open book
Retrieval-Augmented Generation is a fancy phrase for a simple trick: before answering, the AI looks things up in your documents — your company files, your product manual. So instead of answering from memory (and sometimes guessing), it answers like a student allowed to bring the textbook into the exam.
04🔢 Embeddings — turning words into numbers
Computers compare numbers, not meanings. Embedding models turn a sentence into a long list of numbers so that “cheap flight tickets” and “low-cost air travel” end up with nearly the same numbers. This is how AI knows two differently-worded things mean the same thing.
05🔌 MCP — the universal plug
Model Context Protocol is a standard connector that lets an AI safely plug into tools — your files, GitHub, Slack, a database. Before MCP, every connection needed custom wiring. Now it’s like USB: one port design, many devices.
06🛡️ AI Security — the guard
Guardrail tools sit between users and the AI, blocking bad instructions in and bad answers out — leaked personal data, harmful content, or trick prompts designed to make the AI misbehave.
AWS Bedrock Guardrails07📹 Observability — the CCTV
These tools record what the AI did and why: which question came in, what it looked up, what it answered, how much it cost. When something goes wrong, this is how engineers replay the tape.
08💾 Memory — remembering you
By default, an AI forgets you the moment a chat ends. Memory tools store what matters — your name, your preferences, last week’s conversation — so the next chat continues instead of restarting.
09🧰 AI Agent SDKs — the builder’s toolkit
These are ready-made kits that developers use to assemble all of the above into an actual working assistant, without building every piece from scratch.
AWS Bedrock Agents10🗄️ Vector Database — the smart filing cabinet
Those number-versions of your documents (from #4) need to live somewhere searchable. Vector databases store millions of them and find the closest match in milliseconds. It’s the filing cabinet where the open book (#3) does its looking-up.
How they all fit together
In one sentence: your question goes past the guard (6), the planner (2) decides the steps, the brain (1) thinks, the open book (3) checks your files stored as numbers (4) in the filing cabinet (10), tools connect through the plug (5), the CCTV (7) records it all, and memory (8) saves what matters for next time — all assembled with the toolkit (9).
Frequently Asked Questions
Do I need all 10 layers to use AI?
No. If you just chat with ChatGPT or Gemini, you need none of this knowledge. This stack matters when a business builds its own AI assistant — then each layer becomes a decision.
What is the difference between Agentic AI and AI Agent SDKs?
Same idea, different altitude. Agentic AI (layer 2) is the planning concept and the frameworks behind it. AI Agent SDKs (layer 9) are the packaged developer kits for actually shipping one.
Which layer should a beginner learn first?
RAG (layer 3). It is the most-used piece in real business AI, and the easiest to understand: AI plus your own documents. Our guide to how AI search reads content is a good companion read.
