1Mby1M Virtual Accelerator AI Investor Forum: With Naganand Doraswamy, Managing Partner and Founder at Ideaspring Capital (Part 4)

Sramana Mitra: I think for entrepreneurs listening to this conversation, there are two takeaways that you want to consider. One is, find a platform that you can build on. If you’re building enterprise products or mid-market B2B products, look at a platform called Cohere.

Cohere is an LLM. It’s a full-fledged AI platform, but it is focusing entirely on enterprise. It is a B2B enterprise platform. So that may be a place where you will find a lot more infrastructure for building B2B apps. Of course, Microsoft Azure is also supporting B2B app development in a big way. So are Google and IBM. That’s on the platform side.

The second point is the services conversation we just had. If you have domain expertise in a particular domain, whether it’s a vertical or some cybersecurity element or something that you really understand deeply, where AI can make a big difference; go out and do services projects, maybe on top of a platform and start developing solutions and start building IP. It’s OK if it takes you three years to productize, because the beauty of services projects is that money will come in right away.

In services projects, you can design the milestone payments with one payment in advance, and then maybe three other payments. Your seed capital can come from your services projects. So that’s a very powerful way of building products tried and true way. If you’re looking for case studies to refer to bootstrapping using services, our curriculum is a very good place to go look at a lot of curriculum case studies.

AI is also expensive to bootstrap, right? AI engineers cost more money than regular engineers, people who understand AI are at a premium right now. It’s going to be very interesting to be able to leverage the bootstrapping using services model to get at least develop your understanding of what the market needs and how you can monetize as you go along.

Naganand Doraswamy: Another comment also is the fact that the cost of influencing is not cheap. Training costs exist and then influencing costs also exist. It’s all adding up from the cost perspective.

Interestingly, for one of my startups, which has some aspects of behavioral sciences and AI, it’s becoming harder to sell to enterprise. More and more questions are being asked by enterprises when you take a product to sell to them. I think there are a few sets of challenges that have still not come to light, which will need to be figured out. Like we used to have ISO 27001 for security, I think you’ll have a whole bunch of standards coming for AI to minimize hallucination. Prove to me, there’s no bias and all those things.

Sramana Mitra: Hallucination is a very big issue, right? If you’re doing a healthcare app or healthcare solution using AI, and it starts to hallucinate, that’s unacceptable.

Naganand Doraswamy: There’re still a lot of issues that need to be figured out. From what I have heard in the industry and outside, amongst all the tech waves we have had in the past ten years, this is the most promising. So we’ll see.

Sramana Mitra: It is promising. It is a very powerful technology, but I think it’s not going to go as fast as people are projecting or apprehending either, because all these elements have to be worked out, right? An AI application doesn’t get dropped into a vacuum situation. There’s all this workflow around, there’s organization around, there’s integration around, all of that must be figured out to bring an AI solution into any situation.

So now, how do you parse the three years out, five years out, seven years out question? Because as a VC, you’re going to have to deliver fast-paced revenue in that window. So how are you thinking about what the future looks like?

Naganand Doraswamy: We had seen some AI aspects already coming into some of the other companies that we have had in our portfolio, like using deep learning and machine learning. As per a study we conducted recently, in India, 30% of funded companies are still deep learning, machine learning based companies. Not everything is GenAI.

In the enterprises, I fundamentally believe you will get contextual elements and LLMs coming into play, right? I doubt very much a situation where enterprises will use everything that OpenAI brings. We believe that there will be proprietary LLMs in various verticals that come into play. Now, how do you use that and build solutions for enterprises where I think it must become agentic, where tasks can be replaced by workflows?

There are two ways for AI to be useful. In the first way, it produces the first draft of everything, whether it’s code or whether it’s presentation or anything that you want. Then a human intervenes. So, the first draft or version is done by AI.

The second way is where it completely takes over what the human being would be doing, like in contact centers or in testing or whatever. So these are the two areas we are looking at.

We feel more comfortable in in doing number one right now. Agentic, where everything is completely automatic is a little far out. So we’ll focus on the first problem I discussed in our investments right now. That’s what we think will happen in the next three years.

Sramana Mitra: Interesting. You know, I think investing in machine learning companies that do not use generative AI is actually safer because that does not hallucinate. Domain-specific machine learning does not hallucinate. So that’s number one.

Number two is using small vertical language models that are constrained is one way to check hallucination, check bias etc. and create things that are more precise. I kind of agree with you that that’s one way to control the issue of hallucination and create more reliable solutions.

On this first draft of everything point, as a writer, I do not like to use generative AI as a first draft because it has no style, and if you’re trying to do something differentiated in writing, it’s not easy to do that with generative AI. It’s very generic. By definition, it’s very generic. You can ask him to write in somebody’s style and so forth, but it’s not stylish writing. It doesn’t have punch. I think it’s very helpful to get some work done, but I think there’s going to still be room for people doing more interesting copy writing in advertising in media and so on.

So how do you take generic material and then stylize it is a question that people should ask because ultimately if everything sounds like generative AI, then no one’s going to pay any attention to anything, right?

Naganand Doraswamy: So there are two aspects to it, right? One is where, like you say, as a writer, you want to be very creative and those segments might not lend themselves greatly to this. But if you come to the corporate world, whether it’s writing code or whether making presentations or whatever, right?

Sramana Mitra: Code is a very different issue. Yes, code is like, you know, going to get automated.

Naganand Doraswamy: Yes. So code is one example. For an internal corporate presentation, the first draft normally takes time and then you can modify it, right? But Microsoft is doing most of the things now, giving a copilot. I think that helps. So there are certain aspects where you want to maintain individuality, or your own style. It might not lend itself to creativity very well, but in many jobs, first drafts don’t require the kind of creativity that you want, including code. You can get creative, but sometimes you say, don’t be too creative because it will not become maintainable. So in those cases, it becomes a first good option.

Sramana Mitra: Yes, I think coding is going to get automated because that is so deterministic, right? Coding is determined, it’s like math. Coding is going to get automated and that’s that.

Naganand Doraswamy: While looking at a couple of companies, I’m asking myself the question, “What will OpenAI do? Where will OpenAI be in a year? If you give it a requirement document, will it churn out code?” It already can do it.

Srinivas, who is the VP, had come to Bangalore and we were talking about it. He said, “We will substitute many things, but there are certain things that we will not touch. So, as long as you’re in the periphery and doing those things, you’re safe. But figuring out what is that periphery, what is meaningful, where it makes sense for enterprise, I think that that’s the hard part.”

Sramana Mitra: So, you know, I think this is the big danger that is coming is that we have created a society where tech savvy programmers are at the top of the pile. They make a really huge amounts of money right now and they are kind of controlling the prestige and everything. Their job is going to get automated.

So now comes the question, what do you do with millions of tech savvy people who understand AI at a pretty good level?

I think that if we can take millions of people who have some understanding of technology and give them entrepreneurship education, then they can use their technology knowledge and use their entrepreneurship knowledge and actually build companies at a higher level of abstraction. That could be very interesting.

Naganand Doraswamy: It needs to be seen how that will pan out going forward. All I can say is we live in very interesting times.

Sramana Mitra: I think a lot about the layers of abstraction question because when I was in school in the mid-nineties, everything was so expensive, right? A laptop was expensive. None of these layers of abstraction existed. So, today we have AWS, Azure, etc. and many other layers of abstraction. You have the cloud, you can basically provision any amount of capability from the cloud with whatever you need to do. The layers of abstraction have given us ability to bring technology down into the depths of small businesses, right? In the beginning, all this was only available to large companies, large enterprises. And now as a result, entrepreneurship became much faster, much nimbler, etc.

But it’s still very expensive to do an entrepreneurial venture, especially if you have a lot of technology to build. If there are more layers of abstraction that makes that easier, that could unlock another big entrepreneurship renaissance.

But then comes the question, what problems are they solving?

That’s going to be the real question then.

Naganand Doraswamy: I think software development has become more of integration as opposed to writing a lot of code now, because you’re pulling things together and in a certain way where it gives the outcome that you want, as opposed to our days, where we used to write socket code, right? People don’t even know what a socket is anymore.

Everything is given to them. If you want to write any web programming, so much abstraction has happened that you don’t even need to know the basics anymore. The paradigm itself has changed significantly over the past twenty years.

Sramana Mitra: Well, fascinating conversation, Naganand. Thank you for coming today and look forward to, we’ll catch up soon.

This segment is part 4 in the series : 1Mby1M Virtual Accelerator AI Investor Forum: With Naganand Doraswamy, Managing Partner and Founder at Ideaspring Capital
1 2 3 4