The five errors you'll see ninety percent of the time

Python has dozens of built-in error classes. AI-generated code, in practice, hits maybe five of them again and again. If you can recognize those five on sight, you can diagnose almost any crash in the first second of reading the traceback — before you read the message, before you check the line number.

This is one of those force multipliers that turns "AI's code is broken and I have no idea what's wrong" into "oh, that's a KeyError, let me check the JSON shape."

The five core errors and what they actually mean

Error	Plain English	Where it comes from
NameError	A variable name that doesn't exist	Typo in a variable name, forgot an import, used something before defining it
TypeError	The operation doesn't work on this type	Mixed types ("5" + 10), passed None where a value was expected, wrong number of args to a function
KeyError	A dict doesn't have the key you asked for	Typoed key, key missing from API response, capitalization mismatch
IndexError	A list isn't long enough	Off-by-one bug, list came back empty, asked for items[5] on a 3-item list
AttributeError	You called a method on the wrong type	Almost always None.something() — a function returned None and you tried to use it

Read the table once. Come back to it whenever a crash surprises you. Within a week the translation is automatic.

The diagnostic shortcut

Each error class is a fingerprint — a signal of which kind of bug you're looking at, before you even read the message. When AI hands you a stack trace:

• See NameError? It's a typo or a missing import. Look at the exact name in the error and search the file for it.
• See TypeError? Something is the wrong type. Often None showing up where a value should be — which means a function returned None instead of what you expected. Hello, return-vs-print bug from chapter 2.
• See KeyError? A dict is missing a key. Either the key is typoed in the code, or the data shape isn't what AI assumed. Print the dict's keys and compare.
• See IndexError? A list is shorter than expected. Empty list, off-by-one, range mismatch.
• See AttributeError? You called .something() on something that didn't have it. Usually None. Find the function whose return value you used and check why it gave you nothing.

Where AI specifically gets each one wrong

Cursor doesn't just read these errors — it also generates them predictably:

• NameError from refactor drift. AI renames a variable in one spot and forgets the other usage. The script crashes the first time it hits the unrenamed line.
• TypeError from None propagation. AI writes a function with a missing return. Calling code uses the result, blows up on the next operation. Symptom is a TypeError two lines downstream of the actual bug.
• KeyError from optimistic API shape assumptions. AI writes response["data"]["users"] based on the docs — but the live API returned {"error": "rate limited"}. No "data" key, no defensive check, instant crash.
• AttributeError: NoneType has no attribute X. The all-time champion of Python bugs. Some function returned None. Find which one. The fix is usually upstream of where the crash is.

The two extras you'll also see

Two more errors that don't make the top-five but appear weekly:

• ValueError — right type, wrong content. int("hello"). The argument type is correct, but the value can't be converted.
• ZeroDivisionError — divided by zero. Almost always means the denominator came from a count that was empty.

When you see these, the fix is content validation, not type validation. We have a whole lesson on those two coming up next.