I've been building small AI-assisted apps for a while, and I noticed something across the ones that made money and the ones that didn't. The difference almost never came down to the build itself.Two ideas, same weekend, same AI coding agent, roughly the same skill level going in. One founder ends up with paying customers within a few weeks. The other ends up with a working app and zero users, wondering what went wrong.Once you look for it, the pattern is pretty consistent. The people who succeed aren't better at prompting. They're better at knowing which ideas deserve a weekend of their time before they spend it.That's the part that doesn't get talked about much. There's a lot of content out there on how to build with AI tools. That's basically free and everywhere now. There's a lot less on how to tell, in advance, whether the thing you're about to build has any real chance of working.Here's the checklist I run every idea through before I let myself open a coding agent:1. Is the problem specific, or just directionally true?"People want to be more productive" isn't a problem. "Freelancers lose unpaid hours because a client's 'just one small thing' request quietly falls outside the agreed scope" is a problem. If you can't point to a concrete moment where someone hits the pain, it's not specific enough yet.2. Can you name three places your exact audience already gathers?A subreddit, a Discord, a niche newsletter, a hashtag. If you can't name any, you've just found a distribution problem before writing a line of code, which is a lot better than finding it after.3. Is anyone already paying, in any form, to solve a version of this?Could be a mediocre tool people tolerate, a freelancer people hire manually, a course or template people bought about the problem. Zero existing spend isn't automatically fatal, but it means pre-selling should be mandatory for you, not optional.4. Could a competent developer copy this in a weekend?If your honest answer is yes, that's actually useful information. It usually means you need to go deeper into a niche, or add something on top of the AI layer that isn't just a prompt with a UI. A workflow, a data advantage, an existing audience, some real embeddedness in a specific industry.5. Once you have users, how fast can you ship a fix?If ten people tried your MVP tomorrow and five wanted the same specific change, could you ship that in days? If it's more like weeks, your iteration loop is too slow to compete with someone who can turn things around in a day.None of this needs a prototype. All five can be checked in a few days, in parallel, before any code gets written.The uncomfortable part is that most builders skip this. Not because they don't know validation matters, but because building feels like progress in a way that customer interviews and landing page tests don't. Enthusiasm from people you talk to is cheap. Evidence that money is already changing hands, or that a landing page converts at a real rate from targeted traffic, is a completely different signal.I wrote a longer breakdown of this approach, along with a full scoring system for turning the checklist into an actual go or no go decision, in a book called The Moat Test, if it's useful.
0 comments — join the conversation on Findborg