A founder pitch: why most AI agents are built wrong

Transcript

Kind of interested in your product, but I think you have to be a little more gutsy. I'm gonna give you one lakh rupees. What are you building for me? There are a thousand AI agents startups right now, but here is what I think they're all getting down. Companies are not built on code. They're built on people, right? And that is why we are building internet humans. Internet humans are like real humans, but digital. They have their own name, their own phone number, their own email, and their own personality. And yet if companies are the next obvious thing. But again, as I said, companies are built on people not code. This is all possible because we run a tiered memory system with a short term, long term and a permanent memory system. Everything is weighted and we do a cleanup every night. We basically make the model debate on what memory is important and what's not. And lastly, we have a humanizing layer so that we never break the illusion that we are AI. Now the first internet humans are out there as D2C relationship managers where they help in pre and post purchase decisions for all the customers that they have and also complaints and feedbacks, basically driving abandoned carts and articles down while the revenues go on. So you have built this already or you are going to build it. Right. So we have already built this. This is live with two design partners of ours. One is a tech company where they have a lead base of 3000. We are selling a 50,000 summer program for them. Which are the two brands that you have? So we have currently Rubaroo AI and then we have Subdra. Because of this conversation it ends in a 62% higher. High reply rate. Reply rate or purchase rate? Our purchase rate is 28% higher. Reply rate compared to a broadcast message on WhatsApp which you'll get from any XYZ brands. Companies will build their own in weeks. I'm glad you has this and that is why we are very very conscious about our dependability on Claude or Open AI or Gemini and you know being eaten up by any of them. So what we have done is we have refined our agent loop and the core of our internet human so much that even if you put an Opus 4.7 model or a Kimi K 2.6 model, right. The result for our use case is not that far off. You should be playing on uniqueness of technology. There are thousands of Chinese knockoffs of Tesla. Tesla is still valued so much about people give credit to the original idea. I kind of am interested in your product but I think you have to be a little more gutsy so I

Transcribed automatically from the video, so small errors are possible.