An AI assistant can often produce a plausible Python program from a short request. That changes how programming should be taught. It does not make programming understanding obsolete.
Imagine asking for a program that publishes a multilingual community event. The generated code may run immediately and still:
change a supplied name or date;
remove an accent or mishandle a writing system;
confuse capacity with attendance;
expose information that should remain private;
depend on a package that is unavailable or unsafe for the setting;
fail on the second record even though it worked on the first; or
solve an easier problem than the one you actually described.
Fluent-looking output is not evidence of correctness.
Your value is not typing speed¶
Learning Python today is not preparation for a contest against a machine. Your value comes from being able to:
frame the problem — decide what should happen and for whom;
read the proposal — trace what code and data are doing;
test the behavior — compare results with explicit expectations;
diagnose failure — use evidence rather than guesswork;
judge fitness — consider clarity, reliability, privacy, bias, and consequence; and
explain and own the result — communicate what you verified and what remains uncertain.
These capabilities transfer across tools. A particular AI product will change. Your ability to specify, inspect, test, and explain will remain useful.
We will use AI gradually¶
This book does not pretend AI assistants do not exist, and it does not hand your learning over to them. Assistance will expand in stages.
In the protected foundation stage, you first build a personal baseline by reading and modifying small programs without generated answers.
In the bounded tutor stage, AI may ask questions, explain an error, or give a hint without completing the task.
In the pair programmer stage, you may compare or improve a proposal, but you must test it and account for the changes.
In the supervised builder stage, AI may help create a larger component while you retain the specification, evidence, and responsibility.
The point is not to delay useful tools for arbitrary reasons. It is to prevent assistance from removing the very practice needed to evaluate that assistance.
A simple trust rule¶
For any code—written by you, copied from documentation, suggested by a classmate, or generated by AI—ask:
What evidence would make me trust this result for its intended use?
In an early exercise, the evidence may be a prediction and an exact output comparison. Later it may include tests, source documentation, data checks, peer review, performance measurements, and a record of limitations.
You are allowed to enjoy the speed of modern tools. You are also expected to know when speed has carried you in the wrong direction.