Recap // the story so far

From spotting patterns
to writing language

Predict
Machine learning learns patterns from data and returns a number or a label.
fraud scoreAML alertcredit-risk rating
The bank has done this for years.
→
Produce
A large language model learns language and returns brand-new text.
draftsummariseexplainanswer
This is the new thing.
Same family of technology — a different job, and a different risk profile.
Recap // the mental model

The model is the author —
the chat is a character it plays

01
The author
the model

Predicts the next word, over and over. That is all it does — at enormous scale.

02
The character
the chatbot

“A helpful assistant” is a role the author performs. Change the role, change the voice.

03
The desk
the context window

The author can only use what is on the desk right now. Off the desk, it cannot see it.

Today // the next layer

Give the author tools and a loop —
and you get an agent

think → act → observe
model
the author
instructions
a tool
a tool
memory
agent = model + instructions + tools, on a loop
And every tool is a control point — something you can log, gate, approve, audit.
Let’s build one, watch it think, and see what it costs.
← → to move