Somewhere in your notes app is a list with a name like "build ideas." It holds four entries, the oldest from last spring. Every few weeks you reread it, feel the same jolt of conviction, and close the app, and not one entry has shipped or even started.
The list itself is not the problem. The reason nothing ships is that a list is a comfortable place for an idea to hide, because an idea you never build can never fail.
By the end of this chapter you will have chosen one idea and picked its form, and the rest of this part builds on those two decisions.
Three tests filter your list to one
Building even a small thing costs a stack of evenings and weekends. Before you spend any of them, run every idea through three tests, in order. An idea that fails any test goes back on the list; the one that passes all three is your build for the rest of this course.
Test 1: does it solve a problem someone actually has? A specific person has to name the problem, what they do instead, and why that workaround annoys them enough to switch. "Freelancers need better invoicing" fails, because it describes a market rather than a person. "My partner chases three late-paying clients by hand every month and dreads it" passes. The strongest pass of all is when that person is you, because then you are the builder, the first user, and the test market at once.
Test 2: can you say the smallest useful version in one sentence? The sentence covers the smallest version that would help someone on day one, not the whole vision. "A tool that makes team retros better" fails, because it names a category. "A single page where I paste raw retro notes and get back the three recurring complaints, ranked" passes, because you could read it and start building tomorrow. If you cannot get an idea down to one sentence, keep narrowing what it does until you can.
Test 3: are you willing to be its first user? Most builds fail this test without anyone noticing. It ships, nobody opens it, the builder included, and a month later it is archived. "I would use this every week" passes. "I would use this if I were a more organized person" fails, because that person does not exist. Every build we have quietly retired failed this test in hindsight.
Run yours now. Take five minutes before reading on and mark each idea pass or fail against the three tests. Most fail the first or the third. If two survive, take the smaller one, since smaller things ship sooner and teach faster. If none survive, your ideas are early rather than bad, so spend a week noticing what annoys the people around you, then rerun the tests.
Then pick its form: web, mobile, or neither
With one idea chosen, the next decision is its form, and there are three doors. Beginners often pick the wrong one because it feels the most like making a real product.
The web door. A web app lives at a URL, runs in any browser on any device, and updates the moment you push a change, with nothing to install. Running it costs almost nothing at first, as you saw in hosting, renting a computer on the internet, and distribution is a link. The AI tools you will build with are at their strongest here, because web code dominates what they were trained on, and no app store stands between you and shipping. This is the right default for almost every first build.
The mobile door. A native app lives in an app store, has to be installed, and updates only when the user chooses to. Pick this door only for a phone-specific reason: the camera, offline-first use on a trail or a flight, or notifications that must reach someone away from a desk. A real reason has to be worth the costs that come with this door: a review process before every release, the store's cut of anything you charge, and usually two platforms to keep working. If the idea arrived as "an app for X," ask whether a really good web app would do the job, because the answer is usually yes and it saves you months.
The neither door. Beginners forget this door exists, and it deserves respect, because many useful things are not apps at all:
- A script that runs on demand and does one job, like assembling a weekly report.
- A form plus a document, where the form collects requests and you fulfill them by hand.
- A spreadsheet with a few formulas, which has carried real businesses for years.
- An automation that moves information between tools you already use.
The question this door answers is not which form looks most like a real product, but which form tells you soonest whether the idea works. Some of the most useful things we have built for ourselves never left it.
Worked examples: picking the door for real ideas
A meeting-prep one-pager belongs on the web. Minutes before a call, you paste the attendee list and your scattered notes into a page and get back one brief covering what was promised last time, what is still open, and what you want from this meeting. It is used at a desk in a browser that is already open, and you will adjust the brief weekly at first, so instant updates and zero install are exactly what it needs.
A packing-list generator does not need to be an app. Friends keep asking how to pack for trips you have done, so the idea takes a destination, dates, and trip type and returns a packing list. The first version is a form plus a document. The form collects the answers, and you keep one tuned list per trip type, adjusted by hand before sending. If people keep using it and asking for variations you cannot keep up with by hand, that demand is your invitation to the web door, with a far clearer spec in hand.
Still torn? Ask which door ships in a week
If you cannot decide, one question settles it: through which door could you put something a person can actually touch within a week? The web door usually clears that bar for page-based tools, the neither door clears it for almost anything, and the mobile door almost never does, which is most of the reason web is the default. Whichever door you can ship through this week is the one that starts teaching you the soonest, and moving to a different door later is normal, as the packing list shows. Write the door next to the idea with a one-sentence reason, because both lines go straight into your plan when you write your Build Plan and start the build.
Try it now
No setup: The drill is the decision itself and needs only something to write on. Write down your three or four ideas. Run each through the three tests in order, and be strict on the third. Keep the one survivor, pick its door, and finish with the two lines you will carry through this course: the smallest useful version in one sentence, and the door with its reason.
With your tools: Open Claude Code and put the decision in front of it with one boundary. Tell it: "Here is my build idea: [your one-sentence version]. I plan to build it as [your door]. Do not write any code yet. List the biggest risks to this being useful, and describe the smallest version you would ship first." The no-code instruction matters, because these tools start producing files the moment a build is described, and today is for pressure-testing the decision. Push back on generic risks until the list gets specific to your idea. In Codex or Cursor the move is the same: give the sidebar chat your pick and the no-code boundary, then ask for risks and a smallest first version. If your tools are not set up yet, the Setup Clinic gets you there in one sitting.
Chapter Summary
- A list is where ideas go to feel safe. You leave this chapter with one idea you have committed to building.
- Run every idea through three tests in order, and send any idea that fails one back to the list.
- Test one: a specific real person, ideally you, can name the problem and the annoying workaround they would drop for it.
- Test two: you can state the smallest useful version in one sentence you could start building tomorrow.
- Test three: you would honestly use it yourself. Most abandoned builds quietly fail this one.
- Then pick a form. Web is the right default because it is cheap, updates instantly, ships as a link, and is where your AI tools are strongest.
- Choose mobile only for a phone-specific reason like the camera, offline use, or notifications, since it costs you reviews, a store cut, and two platforms.
- Do not forget the neither door: a script, a form plus a document, a spreadsheet, or an automation often proves the idea fastest.
- Still torn? Pick the door you could ship through within a week, and write it beside the idea with a one-sentence reason.
- Next we look at what the model should do inside the thing you picked, in Where AI fits in your product.