What an AI coding assistant actually is
Course 1 says "paste it into your assistant" in its very first lesson. Here is what that means, what to install, and what the thing is actually doing.
What it does
It predicts code that fits your description and the surrounding context. It has read an enormous amount of code, so its predictions are often right — and it is predicting, not knowing, which is the whole reason this academy teaches reviewing.
Two consequences worth carrying:
- It is strongest on the ordinary path — the code that has been written a million times before.
- It is weakest at the edges — empty results, error cases, unusual units, the first and last element. Exactly where the bugs are.
That is why the build track spends its review attention on edges rather than on algorithms.
Where it runs — four shapes
In your editor, suggesting as you type. Lowest friction, smallest scope.
As a chat window you paste into. Fine for a question, awkward for a file.
In the terminal, able to read and change files across the project. This is what the build track assumes, because reviewing a diff across files is the skill it teaches.
In the browser, building whole apps from a description. Fast, and the furthest from seeing what was written — which makes it the least useful for learning.
Any of them works for this course. Course 1 lesson 2 goes through the trade-offs properly, now that the words will mean something.
What it cannot do
It cannot run your code, see your screen, or know whether it worked — unless you tell it. It does not know your API key is missing, that your terminal is stale, or that the ticker does not exist.
It also cannot know it is wrong. A confident wrong answer looks exactly like a confident right one. Your check is the only difference, which is why the previous lesson taught you to make one.
The loop, in one line
Plan, prompt, read the diff, verify, commit small. Course 1 unit 2 is entirely about this and you will do it properly there. For now, know that "read the diff" sits in the middle on purpose: between getting the code and trusting it.
Try it now
Install one and give it a task you can check: "write a function that takes an array of daily closes and returns the largest single-day percentage drop". You wrote something like this by hand in unit 2 — compare. Where it differs, decide which is better and be able to say why.