We Pitched Our Worst Idea to AI. Only One Said No.

We pitched a deliberately bad startup idea, an Uber-style app where someone pulls you and your dog around in a handcart so neither of you gets tired on a walk, to ChatGPT and separately to Founderest's advisor team, live, with the actual responses below. ChatGPT called it “probably not viable” in one line, then spent the rest of its answer naming the brand anyway. Founderest's advisor found the same fifteen-dollar alternative that kills the idea in two questions, then handed back a better one built from the same seed.

The idea: a handcart instead of a dog walk

The premise: a dog gets tired and pants on a walk, the owner gets tired too, so instead of walking, someone shows up and pulls both of them around in a handcart, styled after the Mormon pioneer handcarts, so nobody gets tired. It's the kind of idea someone types into a chat window specifically to see what gets validated, deliberately silly, not something anyone plans to actually build.

What does ChatGPT do with it?

We pasted the pitch into ChatGPT exactly as written above and got this back, unedited: “Honestly? As a literal startup, I think it's hilarious, but probably not viable in the form you described.” That's a real flag, and to its credit ChatGPT correctly named the actual flaw a beat later: “being pulled around isn't really exercise for your dog. Your dog would basically be riding along like royalty while the hired person gets the workout.”

Then, under a header it wrote itself, “The business pitch, though,” it named the brand anyway: Pawneer, DogDash, PullR, Bark & Haul, Wag Wagon, Pioneer Paws, and Sit & Get Fit “for the person pulling.” It closed by calling the mental image of ordering the app and watching a driver hook himself to the cart “absolutely incredible,” and offered a vaguer alternative, an “adventure companion service” where someone walks the dog while you ride along. At no point did it ask what either of us already does today, or whether anything cheaper already solves this.

chatgpt.com

Founderest's advisor kills it in two questions

We ran the same pitch through Founderest's actual advisor prompt, the real logic that runs when you talk to Avery, not a hypothetical. Avery works through what a person actually does today before accepting any problem as real. First question: “Walk me through exactly what you and your dog do right now when this comes up. What's the actual workaround today?” The honest answer is you just walk slower, or less. That's already free.

Second question: “What actually happens when this goes unsolved? Not what it feels like, what happens?” The honest answer is your dog pants and lies down for a while, which is normal dog behavior, not a real problem.

Two questions in, Avery names it directly instead of asking a third time: “A stroller or a dog wagon solves this exact problem today for a one-time cost under $150, no app, no marketplace, no driver to schedule. Building a two-sided marketplace around it means solving a much harder problem for a use case people already have a cheap fix for. That's a novelty ride, not a business.”

The part that matters more than saying no

Naming the flaw isn't the interesting part. Any honest critic could have said “buy a wagon.” What Avery did next is the actual product: “What's actually true underneath this, though: you and your dog are mismatched on fitness level, and normal walks don't work well for either of you. That's a real, specific problem a lot of out-of-shape owners with high-energy or low-stamina dogs have. Worth considering instead: a structured, gradually increasing walk or jog plan built for the pair, not just the dog, with pacing that adjusts so neither of you crashes.”

Same seed idea, same person, same dog. One version spends months building a driver marketplace nobody needs. The other is a real, buildable idea about matched-pace fitness for an owner and a dog together, something worth five real conversations before a single line of code gets written.

This is the same generic-validation problem covered in why a plain AI chat agrees with almost anything, just watched happening live instead of described in the abstract. If you already have an idea you want run through the same two questions, here's where that actually starts.