# whynot

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Matteo Atkinson](https://cerebralvalley.ai/u/Chocolatine75), [Tom xss](https://cerebralvalley.ai/u/xdd)
- **GitHub:** https://github.com/Chocolatine75/whynot
- **Demo video:** https://www.youtube.com/watch?v=dagCy-_bUQo
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3565

whynot is an ad agency for small businesses on Facebook and Instagram, run by AI agents.

We built it because a friend runs a language app that makes about $20k a month. He knows his ads matter, but an agency costs thousands a month, and doing it himself means guessing which ad will work and paying Meta to find out.

With whynot, you paste your App Store link. The agents read your competitors’ running ads in the Meta Ad Library and your app’s reviews, then write a strategy: who your customers are and what stops them from installing. They make static and video ads from that. Your real app screen is cut out of the store screenshots with SAM 3.1, so the product in the ad is never AI-generated.

Before any money is spent, every ad goes into a simulated feed, next to your competitors’ real ads. AI viewers playing your customers say whether they’d stop, whether they believe it, and why not. The agent rewrites the ad to answer that objection, and the new version is kept only if it wins head to head by more than the measured noise.

Then whynot publishes the campaign on your Meta account. Everything is created paused, nothing goes live without your approval, and there’s a hard daily cap and a kill switch. Every day it checks results, pauses tired ads and makes new ones.

We tried to be honest about what works. Every simulated number is labelled as simulated, and we publish our failures: the simulator can’t yet predict which ads run longest. A full campaign takes about 6 minutes and $1.78 in model costs.

Built with Muse Wonder, Muse Spark, Muse Spark Open, Muse Image, SAM 3.1 and Voice Transcribe.

## More from Meta Global AI Developer Hackathon

- [Fab Gym](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3562)
- [RGB Reliable General Business LLC](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3563)
- [Boredom](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3564)
- [FieldProof](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3566)
- [Tren-Tren](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3567)
- [The Beit El Team - Bizscore](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3568)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3565. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt