I built an AI panel that argues with each other, because no single AI Assistant ever told me my idea was bad
I had a small SaaS idea sitting in my notebook for almost 2 years.
Every few months I asked an AI Assistant if it was good. It always said yes.
The problem: AI Assistants are trained to agree with you. Their loss function rewards keeping the conversation going, not telling you the painful thing. Ask the same idea 8 different ways and you get 8 versions of "great idea, here is how to start."
So I built the opposite. Seven specialised agents, each given a different lens (market, GTM, technical, founder fit, competition, timing, financial), each isolated so they cannot see the others' answers. They argue. They disagree. The result is a structured Reality Check that tells you what would kill the idea before you spend 3 weekends on it.
A few principles I am holding myself to:
Calibration in public: every prediction we make gets graded against what actually happened. Posted live. You can see where we have been wrong, by vertical, with numbers.
Citation-backed: every claim in the report links to a source. No "studies show" without the study.
Brutal, not edgy: the goal is to save you 3 weekends, not to roast you for clicks.
Here is a real Reality Check we ran on a B2B vertical SaaS idea (anonymised), verdict "worth a closer look":
https://ceos.ro/r/22b88ac784e743b5?utm_source=indiehackers&utm_campaign=ih_v1&utm_medium=community
If you have an idea that has been sitting in your notebook for a while and you want to know what it is missing, the first one is free, no card:
https://ceos.ro/try?utm_source=indiehackers&utm_campaign=ih_v1&utm_medium=community
Curious what the community thinks: when you "validate" an idea today, what are you actually doing? Asking a friend? Posting on Reddit? Asking an AI? I am collecting honest answers because most "validation" is just permission-seeking in disguise.
