@snehal

3x Entrepreneur, 2x Exits. CPO & Co-founder https://xitter.cf/t.co/619zidyWPw, Angel Investor https://xitter.cf/t.co/y0ysfsYUHi. Founded Mezi (acquired by Amex). Founded Dhingana (acquired by Rdio)

 Palo Alto, CA
zeni.ai
Joined June 2007
  • Tweets4.6K
  • Following1.1K
  • Followers1.4K
93Media
I have spent a decade building AI agents, first in travel, now in finance. My last company, Mezi, was acquired by Amex for $125M. It scaled to 100M users. That taught me what people buy. A hand at their back. I am with you; you are not alone. That is what I build at @zeni
1
1
82
Every vertical AI company is about to be valued on what it owns, not what it rents from the labs. Most founders are still answering the wrong question.
1
48
more replies
By summer of next year, owning models trained on your own work, on an open weight base or on the frontier, will be part of how serious vertical AI companies are judged.
1
6
Frontier models will still take most of the spending, even where they lose most of the routine work. If that second part proves wrong, my view of where value sits needs rewriting.
4
The call, in one picture. By the end of 2027, most decision steps inside defined workflows will run on small decision models. The big models will do the talking. Why I believe it, from our floor at @zeni, is in the thread.
Snehal Shinde@snehalOct 15
Replying to @typesafeai
@typesafeai launched Jev last month, and most of the talk is about speed. Speed is the smaller story. Jev does not write text. It returns decisions, with a confidence score you can act on. That is a different kind of model, and finance runs on work shaped exactly like it.
1
75
Replying to @typesafeai
@typesafeai launched Jev last month, and most of the talk is about speed. Speed is the smaller story. Jev does not write text. It returns decisions, with a confidence score you can act on. That is a different kind of model, and finance runs on work shaped exactly like it.
1
1
135
more replies
Two honest limits. The speed and cost numbers are TypeSafe numbers, not yet proven by outsiders. And Jev returns a decision with no explanation attached. In finance, an unexplained number is a starting point, never the answer.
1
29
My call, October 2026. Within two years, the bounded calls inside finance workflows will mostly run on small decision models built for speed and honest confidence. The large models will do the talking and the explaining. The winners will route well between the two.
15
Scale: close to $50M raised, $20B in transactions processed, more than 1,000 customers. Our agents fully handle more than 90% of expense workflows at Zeni. Anything they are not confident about goes to a human for review before it touches the books.
1
18
And here, I share what building that teaches me, plainly and early, and I stand behind the calls I make.
7
Load more