Konstantine Buhler argues AI agents that actually perform tasks, not just advise, are the next frontier. Expo autonomously runs penetration tests better than top security researchers. Rocks, a sales agent, triples performance when humans stay involved. 2025 will bring swarms of agents working together.
Chapters
Are we still hyped? Konstantin.
Good morning, Caroline, and great to see you. We are definitely in an hype cycle, and yet we're not really focused on the 2024 hype cycle or even 2025. We're focused on the very long term. And artificial intelligence is a very big and very important macro trend, akin really to the Industrial revolution. And so that's a trend we have been investing in for decades and are going to continue investing in for decade, decades to come, regardless of where hype lands.
You wrote a piece here or so ago, really identifying how Gartner, for example, sees hype cycles and this wild enthusiasm. And then suddenly you get into this disappointment, sort of a trough of disillusionment. Where are we in terms of consumer adoption, in terms of people actually putting return on investment? Investment first?
You're exactly right. Let me give a couple examples. AI changes over time. When I was an AI researcher a long time ago, we were barely scratching the surface of natural language processing and natural language understanding and the beginnings of computer vision. 2022, 2023 were really about large language models and answering questions. And really the advent of modern AI, 2024 has been the year of agents. And what an agent is, is an AI that actually completes a task. You ask it to do something and it just doesn't just tell you how to do it, it does it for you. An example of an agent company is our own portfolio company, Expo. Expo was founded by the creator of GitHub, Copilot, and they set out to create an agent for cybersecurity, an extremely true, challenging and important space. And in a few months they were able to develop an AI that can actually outperform some of the best penetration testers in the world. So this AI goes in and it attacks a website with the permission of the website provider, finds vulnerabilities and actually reports it back and says, we have found these vulnerabilities. We've completed this entire task that was 2024 for AI was all about agents. 2025 and beyond likely will be about swarms of agents, networks of agents, them working together and maybe even against each other to some result.
The swarm of agents. I feel there's a swarm of companies talking about agents that we're about to have salesforce numbers after the bell today. And they're all about agent force at the same time. We recently had one of your portfolio companies on Rocks, for example, which is also all about agents. How are, how are we going to see them cross contaminate work against each other or Indeed, Some win and some lose.
Hmm. Absolutely. The reason why people are so excited about agents is because the promise of AI completing tasks, not every task, but many tasks, is so impactful. We really want AI to be able to do the work. You just mentioned Rocks and Rocks is a great example of a company that is an agent, but an agent that actually includes the human in the loop. Rocks is not trying to cut the person out. They're actually trying to say, okay, you are a really good seller. You have really good human relationships with your buyers. Instead of trying to dismediate that, we're going to empower you with an agent. This agent is going to do all the work and research for you. It's going to get you prepared for that next meeting. And an example of them actually putting that through theory into practice is they originally developed a technology that would auto draft emails and they actually had the ability to auto send those emails. But they took that out and they found that when they actually kept the human seller in the loop, performance was triple, three times as high. So there are companies where the agent's going to do the entire task, likely in the cybersecurity space, like with Expo, there are companies where their agent is going to do work for the person in order to augment them, like in the case of Rocks.
Constantine, you're a big thinker and you put out sort of trends and forecasts trend for us. How many agent offerings we really need? How many public and private companies, how many horses such as Rocks can you back?
Well, it's just the beginning. Early days for artificial intelligence. And look, the constraint is really our creativity. The amount of opportunity for artificial intelligence is huge. And I'll give an example. Two areas that are critical to all of us are health care and education. They have been areas where cost has gone up and up and up for decades. And a lot of reason for that cost increase is because of the cost of services right now that I can actually complete some tasks. I really think that I can come in and in the coming decades, not just the coming year, but in the coming decades, be a hero for those industries, drive down costs, and actually help us have a better educated, healthier population. So really, the sky is the limit for artificial intelligence and for AI agents and for networks of AI.
