I got tired of AI agreeing with me, so I built a room that argues back
Every AI assistant I used had the same failure mode on hard calls: I'd describe a decision, it would pick up my framing, and hand back a confident answer built on the assumption I hadn't examined. The fluency made it worse — it sounded like analysis.
So I built Azrivo. You describe a decision — consolidate two suppliers or keep both, add a shift or add a line, build it in-house or buy it — and it convenes the specialists that decision needs. They argue it from opposing positions, on record, and converge on the crux: the one question the call actually turns on. You get a verdict, the alternatives it rejected with the reason each one lost, and real deliverables — memos, scoring models, charts you can hand to someone else.
The design bet is that disagreement is the product. A single answer hides its own assumptions; two specialists who disagree have to expose theirs to win the argument.
Free for life; paid only when usage volume grows. I'm most interested in hearing where the debate feels staged rather than real — that's the thing I'm still tuning.
"A room that argues back" is such a good framing. I made a tiny decision-helper for myself with a bunch of personas that disagree with each other, and the weird part was realizing how much I'd been using AI just to feel validated. How do you keep the arguing useful instead of exhausting? That's the part I never solved.
The idea of structured disagreement is compelling, but I'd keep validating what people ultimately trust Azrivo to deliver. The bigger opportunity may not be better AI debates—it may be giving decision-makers enough confidence that the assumptions behind a recommendation have actually been challenged before they act on it.
This is a genuinely useful idea — most AI tools are tuned to be agreeable by default, so building something that pushes back is a nice contrarian angle. Does it argue based on a fixed persona, or does it adapt its pushback style depending on what you're discussing?
当有人反对你的时候你才能思考和检查自己的观点会有哪些可能的漏洞，这对于我们不断完善我我们的计划有帮助。
On your "where does it feel staged" question — the failure mode I'd watch for is the one you're reacting against in the first place: AI's instinct to agree doesn't disappear just because you've split it into two voices. Given opposite stances, they can still quietly drift to a safe middle and call it consensus. The debates that feel real are usually where the two sides are working from genuinely different facts or priorities, not just opposite positions — and the tell is whether either specialist ever holds its ground when the other makes a good point. One that gracefully folds reads staged; one that won't concede the crux reads real.
Hello, I loved your idea. I often get annoyed by my ai assistant because it is always possitive with me. Sometimes I need to argue to improve my ideas. Thank you for your work, update us please!.