This AI Tells You If Your Startup Idea Sucks

What if your co-founder was brutally honest?
Never got tired.
And was trained on the minds of some of the sharpest minds out there.
(the entrepreneur, investor, and operator-type)

Well, that’s what I tried to build last weekend.

Why I Built It

Let’s be real. Most idea feedback is garbage.

You: “I have a startup idea.”
They: “Nice. Sounds cool.”
Which is code for: “I didn’t understand it, I don’t care, or I’m too polite to push back.”

Here’s what founders actually need:

• Real signal: Is this idea worth anything?

• Sharp critique: Where does it fall apart?

• Concrete next steps: What should I validate or kill?

But here’s what most founders do:

• Pitch their mom

• Post in a Slack group

• Pray

So I built something better.

The Agent I Built in 48 Hours

I call it the Idea Vetting Agent. It’s a GPT-powered tool built inside what I’m (for now) calling Superbrains… my AI-powered, founder-thinking companion.

Here’s how it works:

1. You type or speak your startup idea

2. The agent breaks it down like a VC with a caffeine IV

3. You get:

   • An idea deconstruction (market, clarity, audience, etc.)

   • Honest insights (“this part sucks”)

   • Validation steps

   • Repositioning tips

   • Even a narrative angle to test

It’s not polite.
It’s not motivational.
It’s not ChatGPT with glasses on.

It’s like the founder friend who actually gives a damn… but won’t sugarcoat.

What’s Under the Hood

I didn’t train a model from scratch.
I composed it.

Using my personal vault of essays, threads, and notes from the last two decades — Lenny, Tren, Sarah Tavel, Graham, plus investor decks, teardown notes, and my own founder scars. I built:

• Signals → insight patterns from battle-tested founders

• Filters → strategic checks to vet viability

• Frameworks → to structure the analysis like a good operator would

Each piece was distilled, regrouped, and transformed into strategic modules the GPT could use (without mimicking any one person’s voice.)

The result left me stunned!

The feedback feels synthesized, not stitched.
Like a veteran operator who's been in the room, heard this pitch before, and isn’t afraid to challenge your assumptions.

First Test Prompt I Ran

Here’s the idea I threw at it:

“What do you think of an app that teaches K–12 students math in the most interesting way using a character of their choice—say, Harry Potter or a Disney princess? The app focuses on foundational math concepts and teaches learning as a meta skill.”

Here’s how the agent broke it down:

• Strong emotional hook – kids + characters = instant engagement

• Clear problem – math anxiety is real, and parents want better tools

• Sharp challenge – IP minefield. You can’t just drop Elsa into an app

• Weak spot – crowded distribution channels, no virality engine yet

• Cool angle – make math emotionally sticky through story-based learning

• Strategic punchline – “Feels like a killer idea… if you can dodge Disney’s lawyers.”

The GPT kicked back a full investor-style teardown:

• A 5-point evaluation matrix

• IP risk flags and creative alternatives (AI characters, public domain)

• A killer line: “This shines if it makes math emotionally sticky… what’s your coconut?”

• Suggested validation: build a $50 landing page with just the pitch and see who signs up

And yes, it gave a confidence score too:

Medium–High — tons of parental pull, but hinges on execution, retention, and legal clarity

What I Learned

• Strategic prompts create strategic outputs.
  The deeper the prompt, the deeper the feedback. Vague in → vague out. But with this one? The agent gave me real pushback on IP, distribution, and pedagogy. That’s founder-level critique.

• Emotion + friction = signal.
  The idea sparked a big “hell yes” emotionally (of course, parents want this) — but the friction points (legal, retention, curriculum) exposed the real work ahead. That combo is where opportunity lives.

• The AI sounds sharper when trained on tension.
  Because the agent had to balance education, engagement, and legality, the output wasn’t fluffy. It was layered. Insightful. Almost annoyingly honest. That’s exactly what I want.

What’s Next

This is just Agent #1. I’m building a new one every weekend.

Next up:
→ A storytelling agent that helps founders explain their idea with punch, clarity, and investor-grade resonance.

Eventually, I’ll open a public playground.
But right now, I’m keeping it tight.

Want to Try It?

If you’ve got a weird idea that keeps tapping your shoulder.
If you’re a founder, operator, or ghostwriter who helps others shape ideas.
Or if you just want to test something real before you waste 3 months building it...

Thanks for reading.
More weirdness coming next weekend.

-Karthik
Builder of Superbrains (name TBD)
Author of the One wEiRd Idea newsletter

PS: You In?

I'm turning this into a weekly ritual.
One weird idea → built in 48 hours → tested with founders → shared with you.

If you're into weird, useful, slightly dangerous experiments…
Stay subscribed.
And bring snacks.