Skip to main content

​
What AI actually is

Strip away the branding and the hype. The AI you will use today is a language model, built by one deceptively simple process: read a staggering amount of human writing, and learn to predict the next word. That is the whole machine:
  • It does not look answers up in a database.
  • It does not send your question to a person.
  • It writes one word at a time, each chosen as the most plausible continuation of everything before it.
So why does one trick feel like intelligence? Because predicting the next word well forces you to absorb grammar, facts, logic, and style. Prediction at enormous scale starts to behave like understanding. Don’t take my word for it. For the next minute, you are the model.
Every answer Claude has ever written was produced exactly like this: pick a likely next word, append it, repeat. Billions of times, very fast, with a much bigger vocabulary. Once you see that, the most important idea of this workshop follows on its own: If the model continues from whatever it is given, then what you give it decides what you get back. That is why this course spends more time on context than on clever wording. It is also where we go next.

​
From autocomplete to assistant

If Claude only predicts the next word, why does it feel like talking to someone? Because behind the interface, your whole conversation is one continuous document, and it opens with instructions you never see. Claude is the assistant character written into that document. Every reply is the model continuing the script, one word at a time. Flip the view to see your chat the way the model sees it.
This one picture quietly explains a lot:
  • Claude never gets tired, offended, or bored. Characters don’t.
  • It is exactly as sharp at 2am as at 9am. The document doesn’t know what time it is.
  • Nothing about you carries over between chats unless it gets written back into the document. That last one is worth real money, and it gets its own section later: Context & cost.

​
Brilliant at, terrible at

A predictor trained on everything humans wrote has a specific shape. It is superhuman where the work looks like language, and unreliable where the work needs exactness it cannot check. Set your expectations accordingly. Drag a line from each task to where you think it belongs. Connect all six, then score yourself.
Notice the pattern: everything that routes up is language work, everything that routes down is exactness work. And every weakness on the exactness side has a fix. The fixes are this course: tools for math and fresh facts, context for intent, verification for confidence.

​
Five words that unlock everything

Each of these gets its own lesson later. Catch them now and the rest of the day is easy.
ModelThe trained predictor itself. Different models, different strengths. You pick.
TokenThe word-pieces Claude reads and writes. You pay in these.
Context windowEverything the model can see right now. Finite, and yours to manage.
ToolAnything Claude can call that is not prediction: search, code, your systems.
AgentA model in a loop: plan, act with tools, check the result, repeat.

​
What’s in this section

What AI at work really means

Claude as a work system, not a search box with opinions.

Mental model: brain, hands, and memory

Models predict, tools act, context constrains. The picture behind everything.

The AI fluency loop

Assign, equip, inspect, iterate, ship. The habit that makes output reliable.

When not to use Claude

The jobs where Claude is the wrong tool, and what to reach for instead.

Lab: Start using Claude

Open Claude, learn the chat controls, and run your first prompts across modes, models, files, connectors, and research.