MAKER GULLY
I almost opened the editor again last week for an idea that felt sharp at midnight and soft by morning.
Not a big product. Just another "what if Maker Gully also…" sketch. Auth path. Settings screen. A clever edge case I was already designing in my head. The fun part of building solo is how fast you can start. The expensive part is how long it takes to admit you started on the wrong thing.
I have already written about AI as a compiler when I'm writing code, and about AI when it sits in front of a customer. This is the third job: using AI to pressure-test an idea before it becomes a repo.
The cost is not the code
When you're a team of one, a dead end is not "we pivoted." It is three Tired Tuesdays you do not get back, plus the quiet hit to momentum that makes the next idea harder to start.
The hardest part of a solo workflow is not syntax. It is making sure you are not building a clean solution for a problem nobody is trying to solve. AI will happily help you build the wrong thing faster. Used earlier, it can also help you refuse the wrong thing sooner.
Make it argue with you
If you ask a model "is this a good idea?", it will often act like a polite co-founder. That is useless.
I get more out of it when I force a red team. Something like: here is the concept, act like a cynical user and a blunt reviewer. Give me reasons it fails, reasons someone will not pay, and the ugliest friction in the first five minutes.
I am not looking for permission. I am looking for the sentence that makes my stomach drop a little. Heavy API dependency. A workflow slower than the manual version. A support path I would hate answering at 10pm. If Tap and Call had started as "five products sharing one tag" instead of one clear tap job, that kind of prompt would have saved me a month of polite architecture.
The model can be wrong. That is fine. A wrong sharp objection is still better than three weeks of unexamined optimism.
Ask who gets the support ticket
Before Feeder Rater was a loop I trusted, I should have asked a duller question: what confused email does this create?
For any new idea I now ask the model to list the boring failure modes. Permissions. Guest vs signed-in. "It worked yesterday." "Where did my photo go?" "Why is this score different?" If the imagined inbox looks like a second job, the idea is not ambitious. It is expensive.
Identity Swap stayed weekly on purpose. America 250 stays a moment, not a platform. FlashDrop stays a handoff, not a file system. Those are scope choices. AI is useful when it helps me see the support load while the idea is still words.
Write the landing page before the schema
If I cannot get a headline and three bullets that a stranger understands, I do not understand the product yet.
So before frameworks, I draft the pitch. Sometimes with AI, sometimes without. Then I ask the only honest follow-up: does this solve a problem someone is already trying to fix, or am I inventing a need because the demo would look cool?
Feeder Rater got clearer when the page said "upload a feeder photo" instead of talking about models. Tap and Call got clearer when the homepage stopped trailing the feature list. The pitch is not marketing fluff. It is the first validation gate that does not require a deploy.
Save the energy, not the ego
Validating with AI is not about finding a clever reason to quit. It is about stripping the idea down until what remains is lean enough for one person to own.
Open the editor after the idea has survived a harsh review, a support-load check, and a landing page that does not wiggle. Then the hours you spend are hours that compound.
That's the reality of being a team of one.
#TeamOfOne #BuildInPublic #AI #IndieHackers #SoloFounder