From AI to Action: Building Intelligent Agents | Part 1

In Between the Tech – Part 1
AI Powered Engineering
Guest: Nirman Dave, CEO of Zams

Host:
Welcome to Mouser Electronics' In Between the Tech podcast. Today, we'll delve into the integration of artificial intelligence in design tools. As AI rapidly evolves, its capabilities now extend to facilitating software creation, enhancing design efficiency, and much more. Joining us is, Nirman Dave, CEO at Zams, who will share his insights from working with and building these AI agents.

Thank you for joining us today on In Between the Tech. You’ve been starting companies since you were a teenager. What got you started?

Nirman Dave:
Yeah, well thank you for having me. In terms of how I got started building companies, it was never really something that I wanted to do. It wasn't like I set out to say let's build a company. But right after high school when I graduated, my goal was to study for a lot of entrance exams. So I grew up in India and that's where I was at the time. And so the idea was to study for a lot of entrance exams. I can get into good colleges in India or study for SATs and really see where I land. And I took a gap year because I didn't have a lot of time, said, look, let's take a gap year. I'm going to study this, I'm going to spend a lot of time on this. But something led me off that path. And in the gap year, I ended up going to this conference where I was talking about a lot of different hardware tech that I was building at the time. I was in high school when I built my first multi-touch surface table.

So this is, think of it as a really big kind of iPad that's a table. So it was really fun. It was kind of like an open source style project. It wasn't something very production level. But I went to a conference where I was talking about it and I meet this guy who tells me there's a conference happening in Barcelona, and if you apply with your idea and you get in, they'll fly you for free. So as a high school kid that just graduated, hearing the words, free trip to Barcelona for doing something I love was amazing. So I said, let's go ahead and apply with a bunch of ideas and see which one gets picked up. So at the time, Google Glasses were pretty popular, so we said, we're going to do a workshop to have everyone build their own Google glasses.

It didn't go through, got rejected it. We came up with a bunch of different ideas and one of the ideas that we came up with was a very simple idea, which was to create circuits using pencils because a pencil has graphite and that conducts electricity. So we said, let's do a circuit building workshop for kids and if you get selected, they'll fly us for free. Turns out we did. It was such a simple idea that we got selected and all we had to do is pack up a bunch of pencils, a bunch of paper, a couple of button cells, and we were off to Barcelona. When we went there, the kids were loving it, we're doing these fun little workshops. The kids would make a little guitar out of it or they'd make a little circuit where lights kind of pop up every time. So it was a fun little kind of playful thing for kids.

And eventually we had these parents come up to us and say, how much should we pay for this? My kid loves this, I want to take it home. How much should I pay for this? And we're like, what do you mean pay? Is this a pencil and paper and a couple of button cells? You can just have it for free. We had so many parents come up to us at the time, we said, you know what? Just give us 10 euros. It'll be fun. We'll get a sandwich out of it. And at that day, we actually made a little over 1,000 euros and we knew for the first time we can take something like pencil and paper, make some money off of it and create what is called profit and a business. So that's kind of how I got into building businesses. We came back to India, started a company called Circuit Tricks, branded it really well, had some really good designs on it, and eventually we sold that business to a really large company that, at the time, was also integrating these hardware products to their high school curriculum.

And we ended up partnering up with companies like Intel, Microsoft, Arduino at the time because they wanted to get into high school kids and we were kind of the gateway to get into the middle school kids and then really promote their product upwards as the kids graduated. But that was really my first time building something, scaling it, and really knowing what it's like to build a business. And in that process, I spent a year and a half totally forgot about the entrance exams and I started to continue to do this. So that was my journey into building a business.

Host: 
Tell me about Zams and what led you to creating it. 

Nirman Dave:
Yeah, so when I was building Circuit Tricks, I was very much interested in AI at the time. It was very much up and coming. There was a concept called generative evolutionary programming, which means that you use the way humans have evolved over time and you use that technique to kind of prompt an AI model to really come up with a response, to generate a response. And this was back in 2015 when it was like generative AI was not popular at all. This was very experimental. And I connected with a professor online, his name was Professor Lee Specter. He used to run a gen AI lab out of a small liberal arts college near Boston called Hampshire. And he said, Hey, why don't you come down here? We're going to give you a 100% scholarship. It's going to be fun. And I said, cool, free money and free college, how can I say no to that? So, I ended up in Amherst, which is a small liberal arts kind of town.

And as I was building that out, one thing led to another, I moved to San Francisco to work at a gaming company, and that's a whole different fun story of itself. But I ended up working at a gaming company where I was the only data science person at the time. A lot of the models I built generated really meaningful results for the company that got picked up by New York Times and some of the top publishers in the world. And I was one of the interns that was getting access to the board and of the company and really seeing how businesses are built. That's where the idea started was look as the only machine learning person in an 80-person company, everyone would come up to me and say, hey, can you help me build a model to predict sales, to predict inventory, to predict which gamer is going to be more successful on our products, things like that.

And so that's where the idea really started. And we said, look, why don't we build a software that allows anyone to build their own AI models? And as we started to build these AI models, people said, hey, I don't only want to build predictive models, but I want to take action on these models where they would say, hey, look, I predicted what my sales is going to be next quarter. Great, but can I take some action on it? Or I predicted which customer's going to buy again? Great, can I run an automated campaign on it? So, these actions became very, very important. And today we've created a holistic platform that allows any enterprise to build these AI agents that not only forecast outcomes, but also take actions on those outcomes. And it's really built for very nuanced business operations. So, we're not looking at something very high level like, hey, see who's going to churn and make a prediction, but something super nuanced like, hey, which of my vendor is going to have 50 different contracts and which of my vendor is worth negotiating with versus which one's not? Let's have the AI go to the vendor, negotiate these contracts and get me the best price that I can. So, you have an AI agent that can really transform nuanced business operations, and that's what we do at Zams.

The name Zams actually comes from the word Zero Age Main Sequence, which is the astrophysics term for when a star is first born. And so the way we named it is because a lot of our companies that we work with, they end up seeing a fundamental change in the business. They're not just seeing something small here and there, but they're seeing a fundamental change where they're changing the business kind of strategy, they're operations, things like that. And so we said it has to be something core to starting a new chapter, and that's where the word Zams comes from.

Host: 
Can you talk about more about some of the things you've seen with Zams?

Nirman Dave
We have the privilege of seeing the world's largest non-tech native businesses get transformed with AI. A lot of the AI today is built for tech companies. It's built for companies like Microsofts and Googles and all these larger tech companies of the world and which is great, which is amazing. But when you think of a manufacturing company or when you think of a company that's in agriculture, something totally different than tech, when you put yourself outside of the tech bubble, the question is, is this change? Is this revolution we're seeing in AI even relevant to these guys? And if so, how? How would an agriculture company ever even think about AI? Would that even make sense for them or is it just a cool gimmick? What do they really care about? So, I think that's the kind of companies that we typically work with. Those are the type of businesses that want and know that they can see a lot of value with AI, but want to take those first steps and really transform their businesses with that.

So, one of my favorite examples is we work with an agriculture company that actually uses Zams to predict the health of an animal and the health of the embryos of that animal. And so, when they do breeding, they can really identify which animal is most likely to stay healthy, which one's going to need more attention. Very similarly in the agriculture field itself, we see another company that's using us to identify the potential of the fertilizers that it's producing. So, its farmers have a little dashboard where they can see how well the crop yield is coming up. So, these are the kind of predictions and automations that these guys are doing. And now skipping all the way to manufacturing company, we've seen the suppliers of that manufacturer sending them tons and tons of invoices. They get about 10,000 invoices a month. They have dedicated individuals just trying to scramble through those invoices every day. But they built an AI agent that will automatically review those invoices, go back to the vendor if there's any negotiation required, and it saves them a little around $20,000 every month in just negotiating automatically with those vendors. So, we're seeing a lot of these interesting outcomes. Seen one in the shipping industry. We've seen one in finance. In legal. So really, we've seen so many different outcomes where these businesses that originally aren't the tech-forward business, their products aren't specifically tech, but they can really adopt AI in a way that transforms their business.

Host: 
Where do you see the evolution of AI? What's your short-term and long-term vision there?

Nirman Dave:
AI has historically become more and more autonomous, and it's pushing the boundaries. So, the first set of AI that we saw in the world was descriptive AI, which was this idea that you can take a lot of data and you can just plot it into charts. Then we got predictive AI, which is this idea that you can take a lot of data, plot it, and predict what's going to happen next. The third wave is prescriptive AI, which was it predicts what's going to happen next and then prescribes what you should do. So if I predict, or if the AI predicts that the sales are going to drop, it gives you prescription on exactly what to do, kind of get them back up. Or if they predict that a customer is not going to buy again, it gives you a prescription of exactly what to do.

And now we have a third wave, which is what we call as AI agents, where they're not just prescribing, they're going and doing the task for you. So instead of just prescribing, hey, run this marketing campaign to this customer segment, it's actually going to go ahead and do that, and it's going to show you the results. And that's where we are today with AI agents. And as we move forward, we're going to see a lot of these agentic swarms where these agents interact with each other and really get the job done. But also more importantly, they're actively making these decisions themselves. So instead of you telling the agent, oh, look at this invoice and if you want to negotiate, go ahead and negotiate with a vendor. The agent itself makes a decision of how much to negotiate, which vendor to go to negotiate with, and then just comes back to you and says, you know what? Typically, we get about $10,000 of savings. I got you $50,000 of savings. So, this is the world of autonomous agents. And specifically, in that today, a lot of agents rely on what we call APIs. So, if they want to interact with Gmail, when they want to send out emails, things like that, they're going to use those APIs, they're going to interact with them, but there's a world of apps and services that don't have APIs. 
So, the next kind of future step is to say, how does an agent interact with things that don't have APIs, that don't have public facing APIs that you can use? An example is we work with a logistics company and they have a dashboard where they partner up with companies like Samsung to move a few things around and the dashboard that they have, there's no API for it, it's just built for a few logistics vendors that'll manually go in, punch in the numbers and the details in and kind of move forward. Now, that's okay, but how do you automate something like that? How do you ensure that if there's an action that's required, even in those very nuanced, very kind of discrete silos, how can an agent take that? And that's where the world of model context protocol, MCP, comes up. That really drives that change as well. So that's how we would see the evolution I think.

Host: 
AI is in the headlines every day. There are new platforms, new tools. How do you discern fact from fiction with announcements?  

Nirman Dave:
When it comes to AI, there is what's possible today and what the future can hold tomorrow, and there's a lot that is possible today, but it requires a lot of intricate kind of structuring and moving things around and integrating with a bunch of different apps with some tools. And it does require a little bit of technical knowledge where you're A/B testing what works and what doesn't work. But also a lot of what's possible today is not possible at enterprise scale. Meaning you could build an AI agent that goes through a couple of invoices, but how do you build an AI agent that's kind of scaled to go through, let's say a million invoices every month when you're thinking about enterprise scale, how do you scale these AI agents? That's something that companies like us, Zams, we’re kind of building those infrastructure that allows business to do that, but that's just something that naturally is what we see the next thing that comes up.So that's more like the B2B AI side of things. 

And then there's a little bit more philosophical AI side of things like, oh, what happens when everything's done by AI agents? What are humans going to do? Things like that. And naturally what I see and when I think about those kind of transformations in the world is it's just a different evolution of our job. A great example that I can give is there are two very interesting jobs. There's a job of a data scientist and then there's a job of a content writer, people that would write blog posts, content, things like that that I've really seen this change on. As a data scientist, previously, when you wanted to build out any sort of AI models, you would write long python scripts of data, cleaning, prep in structuring, things like that. Today with the AI agents, instead of writing those long Python scripts, you have to become really good at instructing the agent on what to do. And so while your job fundamentally has stayed the same, you're thinking about the right kind of things, you're still thinking about the same problems. The way you're executing is very different. Instead of writing code, you're writing prompts and similarly when it comes to writing content as a content writer, you're thinking of different types of content pillars, what things fit where, how do you really optimize this content for the best reader? Is this for top of funnel? Is this for middle of funnel? Is this for bottom of funnel? You still think about the same fundamental things as a content writer, but now instead of writing and choosing every single word yourself, you're giving the right kind of instructions to the AI agent. So, this became a pretty common term called prompt engineering, where we really take a lot of this idea of, okay, I'm doing fundamentally the same job, but the way I'm executing is changing. And so, when you think about AI agents in the future, there will be still fundamental jobs that'll continue to exist, but the core method of execution will be very different. I think that's where we really see the world more philosophically going towards as well.

Host:
Interested in more of our conversation with Nirman Dave? Subscribe to Mouser's Empowering Innovation Together technology series. You'll unlock the concluding episode and exclusive access to technical articles, videos, and more. Sign-up today at mouser.com/empowering-innovation.