1Mby1M AI Investor Forum: Shripati Acharya, Managing Partner at Priven Advisors (Part 3)

Sramana Mitra: Interesting. My thesis also is that the defensible businesses are domain knowledge heavy in AI. The layers of abstraction to some extent are already in place, and there’s going to be more. Just to give you a feel for my framework, I look a lot at the parallel with the cloud computing infrastructure that came about over time.

Salesforce announced Force.com around 2005 or so and then the app exchange. So, very, very serious companies like Veeva could bootstrap on top of Force.com and get a product out; and they became massively successful. These are multi-billion dollar public companies today, and there were many such instances that happened.

So, I think that is the direction in which the AI layers of abstraction are going to go. You’re going to have a lot of this stuff available to build on top of. Salesforce has announced Agentforce and the Agentexchange, which is exactly the same idea. They want you to build AI agents on top of their platform, and they’re offering an AI agent marketplace – exactly the same parallel of the Force.com app exchange combination.

The rest of the industry hasn’t quite gotten that far. There will probably be something from Google, there will be something from IBM, there will be something from ServiceNow, something from Microsoft, but it’s not as far along. Salesforce culturally was so far along in this direction; I think they just jumped on it right away and announced that even if they may not have everything ready yet.

So, I think that is the framework we are going towards. This should make it on the application layer. It should be much easier for people with domain knowledge to build really powerful solutions on top of these kinds of infrastructure.

Shripati Acharya: Definitely.

Sramana Mitra: Are you seeing agent companies in the deal flow yet?

Shripati Acharya: Let’s take a step back and define agents because people have different definitions for the term. I’ll just say what my definition is, so as not to confuse our audience out there.

I think about agents as programs that can stitch together a series of tasks to perform somewhat larger, higher level complex tasks. So, individually they might be actually both using AI, getting the prompt, acting on that prompt to do the second set of things autonomously, and at the end delivering some result back.

So, the short answer is that, not yet. The one place where we are seeing agents being used is in the most simplistic or the most straightforward one, which is in customer service pieces, right? You have an inbound call coming in, you have an agent, which is understanding the customer and figuring things out. The second place would be in service businesses – think of as somebody, you have an inbound customer rep, but he’s not really customer service for complaints, but like a receptionist just coming and taking appointments. Think of a dentist office where somebody might be doing it.

These are all situations where having a human resource is difficult, because it’s characterized by high churn and high training roles. Small businesses especially find it difficult to hire and this is where agents could work.

What we are seeing is that solutions have come up and people are beginning to look at it, but it’s not something which has caught on like wildfire. What we are seeing consistently is that the AI companies have a very strong agent tech roadmap.

Sramana Mitra: Yes.

Shripati Acharya: So, I think that it’s a question of just phases in which these are deployed. Customers rightfully will first deploy the basic AI solution wherein it is serving as a co-pilot, for lack of a better word, before the comfort is there for the agent to go ahead and take decisions on its own.

Suppose you are in the business of approving certain workflow, you are in compliance, or you are in any managerial position or in administration or in HR. Initially you might just get, for example, a prompt that says that, “Hey, this is the request.” You would typically approve it, and you say or click, “Yes, I do”.

After some time, certain things are automatically approved, and the exceptions are flagged. I feel that that’s the way agents are going to go. We have seen efforts at agentic solutions come out, for instance, in writing code, right?

Sramana Mitra: Yes.

Shripati Acharya: But the adoption really has been on the co-pilot side a lot more, which puts the programmer firmly in control, right?

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