Skip to content
sAI
An interactive primer

From words
to workers.

A five-act visual explainer of LLMs, LLM Engineering, Agents, Agentic AI, and the ML-vs-AI engineer split. Built for product people, founders, and curious engineers — no math, no jargon-wall.

Start with Act One →About this site
Before we begin · keep this in your head

Two kinds of people build AI today.

🏭
ML Engineer

Trains models from scratch. Data, GPUs, loss curves. Builds the engine.

vs
🚗
AI Engineer

Builds with finished models. Prompts, RAG, agents. Drives the car to deliver value.

Today is mostly about the AI engineer's world — but as you read, ask which side feels like you.

See the full comparison in Act 5 →
The whole idea, in three lines

An LLM is a librarian. An agent is that librarian with hands. Agentic AI is a whole library full of them — collaborating.

The five acts
How to use
  • ▶Click NEXT to step forward
  • ↶Click any stage chip to jump
  • ← →Use arrow keys + SPACE
  • RReset the diagram
What is an LLM?
01Act One

The Librarian who read every book.

An LLM predicts the next token by recalling everything it has ever read.

Open the lesson →6 steps · ~ 4 min
What you'll see inside
  • 01Question
  • 02Tokens
  • 03Recall
  • 04Probabilities
  • 05Pick word
  • 06Repeat
Why LLM Engineering?
02Act Two

The librarian alone isn't enough.

Prompts, RAG, and guardrails turn a brilliant-but-unreliable LLM into a real assistant.

Open the lesson →5 steps · ~ 4 min
What you'll see inside
  • 01The Problems
  • 02Better Prompt
  • 03Add Context (RAG)
  • 04Add Guardrails
  • 05Reliable Output
How LLM Engineering enables Agents
03Act Three

Now give her a goal and let her act.

An agent reuses an engineered LLM and adds tools, memory, and a loop.

Open the lesson →6 steps · ~ 4 min
What you'll see inside
  • 01Goal
  • 02Plan
  • 03Use Tools
  • 04Remember
  • 05Loop
  • 06Done
What is Agentic AI?
04Act Four

Now imagine a whole team of agents.

Many specialised agents collaborating — research, write, review, deliver.

Open the lesson →6 steps · ~ 4 min
What you'll see inside
  • 01Big Goal
  • 02Orchestrator
  • 03Researcher
  • 04Writer
  • 05Critic
  • 06Delivered
Who builds this?
05Act Five

ML Engineer vs AI Engineer.

Two roles, often confused. The ML engineer builds the car. The AI engineer drives it.

Open the lesson →5 steps · ~ 4 min
What you'll see inside
  • 01The Factory
  • 02Build the Car
  • 03Hand It Off
  • 04Deliveries
  • 05Both Win
The whole stack

AI Engineering is the scaffolding.

The five acts aren't separate ideas — they stack. A raw LLM alone is unreliable. LLM engineering makes it usable. Agents sit on top of an engineered LLM. Agentic AI only works because each agent inside it is itself a well-engineered LLM. Pull out any one layer and the whole tower comes down.

Interactive walkthrough
0 / 5
The Stack — value flows from bottom to topClick NEXT to build it up, layer by layer.LAYER 5 · DELIVERED VALUE✨ Real-world product · Real users · Real outcomesLAYER 4 · MANY AGENTS COLLABORATING🔎 Researcher✍️ Writer🧐 Critic📊 AnalystLAYER 3 · ONE AGENT = LAYERS 1+2 + TOOLS + MEMORY + LOOP🤖 Agent — engineered LLM + tools (📅🔍💳) + memory (📒) + loop (↻)LAYER 2 · MAKES IT USABLE📝 Prompts & Examples📚 RAG & Context🛡 Guardrails & EvalsLAYER 1 · THE FOUNDATION🧠 Raw LLM — predicts the next token. Brilliant. Unreliable on its own.
Click NEXT. Watch the AI stack assemble — and see why LLM engineering is the load-bearing layer.

Tip · use ← → keys to step, R to reset

Walk through Act One →
The takeaway

In one breath:

An LLM is a librarian who predicts the next word.

LLM Engineering is everything we add — prompts, context, guardrails — to turn her into a reliable assistant.

An agent is that reliable assistant, given tools and a goal, working in a loop until the job is done.

Agentic AI is a whole TEAM of agents — each a specialist, all collaborating — tackling goals too big for any one of them.

An ML engineer builds the car in the factory. An AI engineer drives it to deliver real value.

Begin with Act One →