When this happens, it's largely because your "communication protocol" isn't aligned.
In this episode, we won't cover new features; we'll talk about how to take fewer detours. I've sorted the high-frequency pitfalls I've hit myself into three, each paired with a remedy that never fails.
Core principle: the two-step method for precisely controlling AI
Before we get into the specific pitfalls, here's a universal communication framework — the two-step method — to help you avoid most communication problems.
Step one: lock the scope. Spell out three things: which component, what to change, and what stays untouched.
Step two: force confirmation. Before it starts, have the AI say how it plans to make the change, and only once you nod does it move.
Sounds like an extra step? But it can save you at least two rounds of rework.
Tip 1: Lock it down precisely — make the AI change only what you want changed
Have you ever run into this: you want the AI to tweak a button color, and the whole page's color scheme changes; you want it to adjust one card's font, and every card's font changes along with it.
The AI doesn't widen the scope on purpose — when it sees the words "change the style," it automatically reads that as a global operation.
How do you avoid it?
Use scope locking to nail the instruction down:
"Only change the button color on the stroke-training card; leave every other card's style untouched."
Once you say this, the AI's understanding is boxed in — only that card changes, nothing else. If you're not sure it understood, add a line:
"The change you described is this: [repeat it back] — confirmed?"
The essence of the two-step method: the less room you give, the less latitude the AI has.
Tip 2: Force confirmation — make the AI say how it plans to proceed first
Many people (me included) start out thinking: "I'll just give a rough direction; the AI is so smart, it should be able to guess what I want."
And the result? It guessed — and guessed wrong.
The root problem: when an instruction is vague, the AI won't proactively ask "what exactly do you mean?" It picks the most likely interpretation and starts working.
How do you avoid it?
Use two methods together:
Method one: lock the scope. Replace vague words with concrete descriptions.
Instead of: "make the card look nicer"
say: "round the card's corners, add two layers of shadow, use a warm tone for the background"
Method two: force confirmation. Before letting it start, ask it:
"Here's your understanding of the requirement: [restate it] — wait for my confirmation before you change anything."
While restating, the AI will spot contradictions or vague spots on its own and proactively ask you to confirm. It costs only a few seconds, but it saves you the back-and-forth edits later.
The essence: better to ask one extra question than to guess and then have to fix it.
Tip 3: Stay patient — wait until the AI finishes before touching files
I came across this one by accident.
While the AI is running a code-generation or UI-modification task, if you delete a file from the project in your file manager at the same time — you think it's deleted, but once the AI finishes, that file "comes back," contents unchanged.
The reason is probably this: while running, the AI keeps a temporary snapshot of the file state; your deletion and its state are out of sync, and when it's done it restores the snapshot.
How do you avoid it?
Just one word — wait.
Wait until the AI finishes the current task and outputs its result, then touch the files. Make it a habit: while the AI is "thinking" or "generating," don't touch anything in the project folder.
A note: if you delete a file and it comes back, don't doubt yourself — the timing was just off; wait for the AI to stop and delete it again.
Today's summary
Three tips and one core principle:
Tip | Remedy |
Lock it down precisely | Scope locking: which, how to change, what stays untouched |
Force confirmation | Scope locking + force confirmation, two steps |
Stay patient | Wait for the AI to stop before acting |
The two-step mnemonic:
Lock the scope down; confirm before executing.
Master this communication framework and your collaboration with AI Studio will go much more smoothly — not because the AI got smarter, but because you learned how to make yourself understood.