1Mby1M AI Investor Forum: Shripati Acharya, Managing Partner at Priven Advisors (Part 5)

The people who understand those enterprise systems the best are in the best position to build those agents. Is there going to be a big opportunity to build agents that are from entrepreneurs?

How do you think about it?

Shripati Acharya: I think this really comes back to a question about how the incumbent versus the new entrant plays out in this. In one dimension, I think that it’s going to favor the incumbents, but in another one, probably not. Clearly, the incumbents have a distribution and they have the relationships, and that is not to be underestimated. Microsoft leads Google by leaps and bounds in enterprise because that’s a deep relationship they have had. The trust that enterprises have in a Microsoft stack of software has been around since the times of Windows; and it continues on to Microsoft 365 and Microsoft co-pilot.

It’s hard to break into these things, because enterprises are not looking for the cheapest stuff, but they’re looking for the lowest total cost of their solution.

I think that’s correct, absolutely that things like Salesforce and Oracle have a very compelling distribution advantage. I think the data advantage is actually lesser than what I thought it would initially be. AI is built on data and one would say that, “Hey, look, these are the guys who are existing and all this data is available.” Now, if you’re building something on it, clearly the person who has the most amount of this data is going to be able to create an AI system, which is going to be leaps and bounds ahead of anything, which is coming from a new entrant.

But what has happened is that we are finding now that synthetic data is very, very good.

Sramana Mitra: It’s very good, Yes.

Shripati Acharya: And synthetic data requires only a reasonably small sample of the real data. Synthetic data trained systems are able to be just very good at understanding all kinds of things. Just to give an example, there are certain things which occur at a very small percentage of the overall sample size, for example, fraud in a financial system.

Suppose you want to build a system, which is very good at detecting fraud. Just making it sit and watching the regular stream of data is not going to actually make it very good at detecting fraud, because fraud is maybe, picking a number, 0.1% of the total traffic. However, synthetic data can amplify that and help you train that system on fraud and have a fraud detection system.

I feel that we underestimate the ability of startups to actually create equally good and better AI systems. I feel that the incumbents have the very traditional innovators’ dilemma issue because the smartest and the brightest AI engineers are not joining these larger companies. They’re going to be starting their own stuff, number one.

Number two, the underlying stack is moving so quickly that the larger companies, by definition, cannot keep just going and jumping off to the latest stuff because their customers will go dizzy.

So, that actually favors the startups.

But like any other disruption, Sramana, I think that startups will have to go without their adopters. They have to go with their universities and folks that agree to trial your systems.

This segment is part 5 in the series : 1Mby1M AI Investor Forum: Shripati Acharya, Managing Partner at Priven Advisors
1 2 3 4 5