You let your AI write the code, and most days you never open a file. But sometimes you want to look yourself: to settle a stubborn bug, to check something sensitive, or to judge whether the work is any good. This part is optional. If you just want to ship, skip it with a clear conscience. If you are curious, or newer to this and want to learn, it hands you enough to look under the hood without turning reading code into a second job.
4.1.1You don't have to read code, but you can
Reading is a tool you pick up for a reason, not a duty you owe every file. Three moments make it worth the minutes:
- A bug the agent keeps missing. When the agent circles the same fix and never lands it, your own eyes on the code break the loop.
- A sensitive area. Payments, login, and anything touching user data are worth reading before you trust them, because the cost of a quiet mistake is high.
- Judging quality. When you want to know if the work is solid or just working, you have to look.
Outside those, let the agent write and move on.
4.1.2What code is made of
Code is built from a handful of parts, and naming them is most of the battle:
- A file is one document of code, like one page in a binder.
- A function is a named block that does one job when you call it, like
sendEmail. - A variable is a labeled box holding one value, like
total = 42. - A class is a blueprint for a kind of thing, say a
User. - An object is one real thing built from that blueprint, one actual user.
That is enough vocabulary to follow along. You are recognizing shapes, not memorizing a language.
4.1.3Follow one path, not the whole file
Never read a file top to bottom. Pick one action, a user logging in, and follow only the functions it actually touches:
- Start where it begins: the login button or the request that fires.
- Step into each function it calls, one at a time, ignoring everything around them.
- Stop when you reach the answer, the check that passed or the value that was wrong.
Everything else on the page is noise for this question. You are tracing one thread through the cloth, not reading the whole cloth.
4.1.4Let the agent be your guide
You do not read alone. The same agent that wrote the code will explain any file, function, or single line in plain words, pitched at exactly how much you already know. Ask it to walk you through the one path you care about, and to stop and define any term you do not recognize.
Do this now: open one file your agent wrote, name a single action in it, and paste the prompt so the agent walks you down that one path in plain words.
