Type hints

Type hints say what type a variable, parameter or return value should be:

python

def search(books: list[Book], query: str) -> list[Book]: ...

ages: dict[str, int] = {"Asha": 22}

point: tuple[float, float] = (12.9, 77.5)

maybe_name: str | None = None           # "a str, or None" (older code writes Optional[str])

Python does not enforce hints at runtime. search(42, 99) still runs, and probably crashes somewhere inside. So why bother?

Your editor uses them for autocomplete and underlines mistakes before you run anything.
Type checkers like mypy or pyright check a whole codebase for mismatches.
Libraries read them. Pydantic validates data from them, FastAPI builds API docs from them, and in Week 13 your function signatures become the JSON schema the LLM sees when it calls your tools.

The typing toolbox you'll meet in AI code

python

from typing import Literal, TypedDict, Any, Callable

Category = Literal["food", "travel", "rent", "other"]   # only these exact strings are allowed

def add(amount: float, category: Category) -> None: ...

class AgentState(TypedDict):                  # a dict with known keys and value types

    messages: list[str]

    attempts: int

state: AgentState = {"messages": [], "attempts": 0}

def apply(fn: Callable[[int], int], x: int) -> int:   # a function taking int and returning int

    return fn(x)

payload: dict[str, Any] = {"model": "x", "temperature": 0.2}   # Any = "could be anything"

Literal shows up in LLM tool definitions ("status must be one of these").
TypedDict is exactly how LangGraph defines graph state in Week 15. Look out for it.
Don't overuse Any, because it switches type checking off.

Try it

Add full type hints to every function in your Week 1 expense tracker. Then install Pylance in VS Code (it comes with the Python extension), set Type Checking Mode to basic, and fix anything it underlines.