{}Route To AI
Courses
Lessons and outlineLesson 1 of 11

AI Concepts in PicturesLesson 1

18 min read

The Big Map of AI: Six Layers

A single map that shows where every AI idea fits, and how to use it to find what to fix.

In plain English

AI ideas sit on six layers: the model, teaching it, senses, action, safety and running it. Think of building a new employee. When something goes wrong, ask which layer is weak, and fix that one.

Layer
One level of the AI stack, such as the model or the safety checks.
Model
The AI brain that reads text and predicts what comes next.
Stack
All the layers that together make an AI product.
Picture it
The six layers of AI
6. Run it for real
Which model, where, what it costs
5. Keep it honest and safe
Tests, checks, review
4. Let it act
Tools, MCP, agents
3. Give it senses
Text, images, voice, video
2. Teach it your way
Prompts, documents, tuning
1. The model
Tokens, attention, training
Try it yourself
Which layer is it?

Place each idea on its layer of the map.

  • Tokens and attention.

  • Search your documents before answering.

  • Reading a photo of a receipt.

  • Calling a calendar tool.

  • Blocking harmful replies.

  • Choosing a small model to save cost.

0 of 6 right

AI can feel like a pile of buzzwords. Tokens. RAG. Agents. MCP. Fine-tuning. Guardrails. It is easier when you see where each word lives. This course gives you a map. Every idea sits on one of six layers. Once you know the layers, you can place any new term in seconds. This first lesson shows the whole map.

The six layers

  • Layer 1, the model: the brain. It reads text and predicts what comes next.
  • Layer 2, teach it your way: prompts, your documents and extra training.
  • Layer 3, give it senses: text, pictures, voice and video.
  • Layer 4, let it act: tools, plugs between systems and agents.
  • Layer 5, keep it honest and safe: tests, checks and human review.
  • Layer 6, run it for real: which model, where it runs and what it costs.

Think of building a new employee. First you hire a smart person, the model. You train them on your way of working. You give them eyes and ears. You give them tools and a role. You check their work. Then you decide how to run the team and pay for it.

What lives on each layer

the-map.txt
LAYER 1  THE MODEL        tokens, attention, context window, training, fine-tuning, LoRA, distillation
LAYER 2  TEACH IT         prompts, context, RAG (search your documents), skills, fine-tuning
LAYER 3  SENSES           text, images, voice, video
LAYER 4  ACT              tools, function calling, MCP, A2A, agents, computer use
LAYER 5  KEEP IT HONEST   evals, guardrails, red teaming, tracing, human review
LAYER 6  RUN IT           open or closed models, small and local models, cost, caching

How to use the map when something goes wrong

Most AI problems have one layer behind them. Ask which layer is weak. Then fix that layer, not all of them.

  • The answer has wrong facts. Layer 2: give it your documents, or better instructions.
  • It cannot read an image or a voice note. Layer 3: use a model that can.
  • It knows the answer but cannot do the job. Layer 4: give it a tool.
  • It does harmful or silly things. Layer 5: add checks and limits.
  • It is too slow or too costly. Layer 6: pick a smaller model, or cache.
  • It reasons badly on hard problems. Layer 1: use a stronger model or let it think longer.

A real product on the map

Take a bank chatbot. The model is the brain. The bank's rules and help pages are layer 2. It may read a photo of a cheque, which is layer 3. It can look up a balance with a tool, which is layer 4. Layer 5 stops it from giving advice it should not give. Layer 6 decides which model runs it and what each chat costs. Every product is a stack like this.

About this course

Each lesson covers one layer, with a picture. You do not need to code. You only need curiosity. The big map is also on its own page, with every idea linked to a lesson.

→Keep the six layers on a sticky note. Whenever you meet a new AI word, write which layer it belongs to.
⚠The map is a way to think, not a law. Real systems blur the lines. Use it to start a question, not to end it.
Key takeaways
  • ✓AI ideas sit on six layers: the model, teaching it, senses, action, safety and running it.
  • ✓Think of building a new employee: hire, train, equip, give tools, check and manage.
  • ✓When something goes wrong, ask which layer is weak and fix that layer.
  • ✓Any AI product is a stack of these layers.
Hands-on exercise

Map a product you use

  1. Pick an AI product you know, such as a chatbot or a voice assistant.
  2. Write one line for each of the six layers: what does it do there?
  3. Mark the layer you think is weakest, and say why.
  4. Write one fix for that weak layer.
Check yourself
Q1A support bot invents a refund policy. Which layer would you look at first?show answer

Layer 2 and layer 5. Give it the real policy to read, and add a check that blocks answers that are not in the policy.

Q2What is the difference between layer 4 and layer 5?show answer

Layer 4 lets the AI do things, like call a tool. Layer 5 checks and limits what it does.

Q3Why use a map at all?show answer

It lets you place any new term fast and find which layer to fix.

Up next
Lesson 2: Inside a Model: Tokens, Attention and Context
Course overview
0 / 11 finished0%