Before you build anything in 3B, it pays to know exactly what you’re building. This is where that starts.
New to the series? Start here.
Here’s the thing about 3B: it builds what you describe, and it builds it well. Which is wonderful right up until the moment you realize you described the wrong thing. The result is polished, it works, it’s just not what you had in your head, and there’s a particular sinking feeling that comes with it. Nearly every time, that feeling traces back to one place: the very first prompt.
Your first prompt sets the direction for everything after it. Give 3B a clear outcome and it has something real to aim at. Give it a vague one and it does the reasonable thing, which is fill in the blanks with sensible assumptions, and now you’re the one untangling a build that was pointed slightly wrong from the start.
I get the temptation to just start typing. It feels faster, and sometimes you really are working out what you want as you go. But a prompt like “build me something to handle equipment requests” leaves almost everything unsaid. Is it an app people submit to? A report? Something that runs on its own? Who uses it, and what should happen to a request once it arrives? 3B will make a sensible call on all of it, and it’ll build that cleanly. The catch is that its sensible call and yours might be two different apps.
None of this means writing a spec before you’re allowed to have fun. It just means knowing roughly where you’re headed before you hit send.
What a clear outcome looks like
A clear outcome covers three things: what goes in, what comes out, and what happens in between. Not every detail, just the shape.
Here’s the same idea, vague and clear:
- Vague: “Build something for equipment requests.”
- Clear: “Build an app where employees submit equipment requests through a form, IT sees them in a list they can filter by status, and each new request posts a short summary to the #it-requests Slack channel.”
The second one hands 3B the shape of the app, the details that matter, and what to do with a request, so it can go build the thing you’re actually picturing. The first leaves all of that open, and open questions get answered by whoever’s holding the keyboard, which in this case isn’t you.
How to define your outcome before building
Before you write the prompt, it’s worth getting clear on a few things for yourself:
- The starting point. How does this kick off, or how does someone use it? A form they fill in, a page they open, a schedule, an incoming message.
- The inputs. What are you working with, and where does it live?
- The result. What should exist afterward that didn’t before? A saved request, a list someone can act on, a message, an updated record.
- The done condition. How will you actually know it worked?
Answer those four and the prompt mostly writes itself. You’ve stopped describing a wish and started describing a job.
Let Plan mode help you think it through
Here’s the part that takes the pressure off: you don’t have to figure all this out alone in your head before you start building. That’s what Plan mode is for. You talk through what you want, and 3B helps you shape it into a plan you can look over and adjust before a single thing gets built. It’s the same planning this whole article is about, except you’ve got company. Reach for it when the idea’s still fuzzy, or when the build’s big enough that you’d rather agree on the shape before committing to it.
The cost of skipping this
Skipping the planning rarely blows up in an obvious way, which is exactly what makes it sneaky. What you get instead is a build that mostly works. It runs, it looks fine, and then you notice the new requests are being flagged somewhere nobody ever looks, because you asked for the flag but never said where. Now you’re reverse-engineering something you never fully described, which is a harder job than describing it would have been.
A few minutes spent deciding what you actually want is some of the cheapest, highest-return time you’ll ever spend in 3B. Spend it before the first prompt, not after the fifth.
What’s next: Mirror your manual steps, for when you know the task cold but can’t for the life of you put it into words.
