B (5:29)

That's awesome. Thank you again Avram, for really impacting gen AI over the tenure of your career. Obviously that pre gen AI period which you talked about and then the post gen which is obviously a lot faster to get answers. But what I really liked is the analogy used with Ironman and Jarvis, right? Agents now can make decisions and we talked about this on the podcast recently as how you can have your own personal assistant now, your pocket, literally. And you can level the playing field with all these big corporations that have deep pockets that are investing a lot in this technology. So thank you. Abram, the next question I have for you, what is the next big leap in AI driven optimization?

A (6:07)

Well, I think it's going to be having those AIs that actually deeply understand your business. Let's say, for example, I've got a factory, right, and my machines might go down and I don't want those machines to go down. So I actually want an AI that deeply understands all the details of my machines and when they see the temperature or the pressure going up to a certain level, knows historically that that's a potential predictor, sends a message to an engineer, that engineer then is also sent the details of how to do the actual maintenance and then goes off and does that. And that machine never fails. That's a big advance, right, that that LLM having not just world knowledge, but actually the capabilities and understanding of what drives the real world and specifically my business, that's going to be a big deal. The same thing might happen when I want to try and increase revenue. If that AI understands which of my customers typically buy what products and knows that, you know, middle aged men over the age of 30 tend to buy organic goods on a Thursday and it's a good time to give promotions, that's really useful information. But if you multiply that by a thousand or ten thousand different micro segments, all of which the AI understands and all of which it's sending millions of messages and promotions around automatically, that's a game changer.

B (7:17)

It absolutely is. And I love how you talked a little bit about how you can have your AI deeply understand your business or your machines, keep ahead of the business changes or maintenance needs. But I really love the fact that these LLMs are now going to be really up to the minute, up to date, if not seconds, and be able to help us be more dynamic in our decision making when we need to be. So appreciate that. Avram, why is agentic AI gaining momentum as an industry classification?

A (7:44)

It's a good question and I think when I look back at it when, when Geni first started as a tool we'd ask it questions and although the first version showed promise, they also made a lot of mistakes and had hallucinations and there was a bit of skeptic skepticism. But the real momentum, which is really what you're asking, is happening because the quality of the LLMs is advancing at a shocking speed. It's now got to the point where it has incredible knowledge, intuition and reasoning. And so what we've now got is the ubiquitous availability of LLMs to the entire world, which means there's a very low bar for creating agencies. If you have an idea for an agent, some basic technical skills and some time, you can create an agent fairly easily. And pretty soon a simple natural language instruction to an LLM will likely allow anyone to create an agent which is then usable. And this reminds me a lot of the rapid emergence of cell phone app ecosystem after Apple launched the App Store. I don't know if you remember what it was like, but what kind of applications were on our pre smartphone phones feature phones. We went from a situation where, you know, I used to work with startups that tried to convince Nokia or telecoms carriers to add their Apple game to a phone and it was almost impossible. And then after the launch of the App Store it became really easy to build phone apps and which used all the special features of a smartphone. And I'll come to the analogy in a second. But if you look at the smartphone, it has compute, memory, portability, location awareness, cameras, web browsing. You know, at Launch they had 500 apps. Within one year there were more than 80,000 apps. And after five years, more than 1 million. Let me ask you, how many different apps on your cell phone do you think you could probably use today? And we don't even think about it. It's completely second nature. And all of these apps are just. Despite the fact creating a mobile app actually took real investment, I'm seeking alpha. In the early days of these apps, we created both an iPhone and Google app. It was a lot of effort, but we needed to do it and it was really important for our customers. So now let's look at what LLMs can do. LLMs are really smart. They have reasoning, they can make decisions. And so what we should expect is just a multitude of AI agents which are created to be able to do the things we want to do. If we want to book theory, there'll be an agent for that. I want to pay my bills, do my taxes, answer questions from my customers. The barriers to building agents are getting lower and lower. And so we should see this massive momentum. And I think it's not an exaggeration to say that pretty soon all of us will be using agents which are built to solve our personal problem need, just like we use phone apps.

B (10:02)

That's awesome. And I love the analogy, you know, with the apps versus the agents. Now we are leapfrogging and that real momentum is happening right now with LLMs, as you've mentioned. But yeah, I can't believe just in the last probably 18 months, two years, I've seen some incredible things as far as leapfrogging and some of the advanced reasoning available today within the LLM spectrum. Appreciate that. And Avram, last question of the day. Can you give a real world example of always optimize with AI at work?

A (10:30)

Sure. So we build this always optimized capability, which is the ability to use AI constantly in the background to analyze your data and then understand what's driving problems in your, in your work environment, and then suggest actions which can then be done automatically by agents if they are available. And so what I'd like to do is actually start by looking at what it looks like if you don't have always optimized. So let's use a typical example of something I mentioned earlier, which is let's say a business like a telco or a bank is seeing increased customer churn. What do they Do? Well, probably a dashboard is sharing churns going up and then when the data scientist team or analyst team has time, they'll do some analysis to try and discover some insights. And when they find these insights why their churn might be increasing, then they'll send them to the marketing team will then review the insights and propose actions. And then probably different people on the marketing team will try and send emails to prevent churn or put some promotions out there where they try and win back the customers. And it's how Princess is lengthy and infrequent and expensive. And it reminds me how Winston Churchill used to describe democracy as the worst form of government, except for all other types of governments which have been tried. So this is a terrible way to optimize your business, but without AI, it's actually the best and most practical way to do that. So now let's look at the always optimized version of this. So RAI would divide your customer base into micro segments of users with similar characteristics like location, age, behavior, similar to what a data science team or analyst might do. But this would be done systematically, in real time in the background. And the insights are then automatically derived as to which customer group is leaving and suggestions as to why that might be happening. You then take these insights and give them to an LLM which very rapidly then becomes an expert on what's driving customers to lead the business. And then it can propose custom emails to be sent to each one of your customers in order to reduce the churn. And in the future with agents, once they could be trusted, these emails can actually be sent directly by the AI. Right now these would generally need to be approved by a marketing team. And this entire process can happen as frequently as we see the data changing without delays, no waiting for people to be available or to have the time to do the work. And so we see this as being an always optimized by AI cycle. And the same thing can happen with fraud detection on transactions in banks. The same thing would happen would be cost analysis. You know, as costs increase, you can immediately be doing the analysis and be proposing actions to be taken. You might want to do energy optimization, there's many, many different applications, but always optimized means that AI does the hard work. And really the key to all of this is that we figured out the way to educate an LLM how to become an expert in the relevant parts of your business. And once you've done that, it's much easier because LLMs are awesome. They've got the reasoning capabilities and the understanding of the world so they can propose actions and then actions can be taken. So we think that actually this whole always optimized capability is going to be perfectly natural for organizations in the coming years as we grow to trust LLMs and generative AI much more.

B (13:18)

Thank you. I appreciate you unpacking that. It's important for our audience as well here. You know using that always optimized in your everyday work doing the hard work right versus the old way where you had multiple silo teams that needed multiple approvals and there was always these dependencies. Now I like how always optimized is where it can go ahead and do some of those tasks almost simultaneously as you start to see maybe customer churn or disruption or something going wrong it can actually do that work real time which I think is really amazing. So Avram it was such a pleasure having you on today and I look forward to speaking with you real soon.

A (13:53)

Thanks so much. Really a pleasure to be here.

B (13:55)

Bye for now.