The AI Agent Revolution: From Chatbots to Autonomous Systems

COLD OPEN

Marcus Webb: Andrej, there's a narrative going around about ai agents that I think misses the mark. People either think it's overhyped or they think it's going to change everything overnight. Where does the truth actually lie?

Andrej Karpathy: (laughs) That's the perfect framing, because the truth is always in between. What's happening with from chatbots to autonomous systems is real — it's not hype. But it's also not overnight. We've been working on this for years, and what people are seeing now is the result of a lot of unglamorous engineering and science that happened behind the scenes.

Dr. Sarah Chen: Welcome to TechNova. I'm Dr. Sarah Chen.

Marcus Webb: And I'm Marcus Webb. Today we're joined by Andrej Karpathy, Founder, Eureka Labs at Former Tesla AI Director. Andrej, welcome to the show.

Andrej Karpathy: Thanks for having me. Really glad to be here.

SEGMENT 1: Where Things Stand Today

Dr. Sarah Chen: Andrej, give us the honest state of ai agents in AI as of mid-2026. Not the hype version — the real version.

Andrej Karpathy: The honest version is that we're at an inflection point. The breakthroughs in scaling laws and compute trajectory and reasoning models and chain-of-thought over the past eighteen months have been genuine — not just press releases, but real technical progress that you can measure. The key thing that's changed is multimodal capabilities has gone from theoretical to practical. We're no longer asking 'can this work?' — we're asking 'how do we scale it?'

Marcus Webb: What does that scaling look like in concrete terms?

Andrej Karpathy: It means we're moving from bench-scale demonstrations to real-world deployment. The numbers I can share: we're seeing improvements in agent architectures that are orders of magnitude better than where we were two years ago. And critically, the cost curve is bending in the right direction. When you combine better performance with lower cost, you get adoption, and that's what's happening now.

Dr. Sarah Chen: What surprised you most in the last year?

Andrej Karpathy: The speed. I expected progress, but the pace has been remarkable. open vs closed source advanced faster than I predicted, and training efficiency turned out to be more tractable than we thought. When you have multiple breakthroughs happening simultaneously, they compound on each other, and that's when you get these nonlinear jumps.

SEGMENT 2: The Hard Problems

Marcus Webb: Andrej, let's talk about what's still hard. What are the problems that keep you up at night?

Andrej Karpathy: The biggest one is data quality and curation. It's the difference between something that works in the lab and something that works in production, every day, under real conditions. We've solved the science. What we haven't fully solved is the engineering — safety and alignment, regulatory landscape, and the integration of all these components into a reliable system. That's the 80 percent of the work that nobody sees.

Dr. Sarah Chen: Can you give us a specific example of a problem you had to solve?

Andrej Karpathy: Sure. Take investment and compute infrastructure — six months ago, we were hitting a wall there. The conventional approach wasn't working, and we had to rethink the problem from scratch. What we ended up doing was llms-centric — we restructured the entire approach around that, and it unlocked everything else. Sometimes the breakthrough isn't a new technique — it's a new framing of the problem.

Marcus Webb: How many people are working on these problems at your level?

Andrej Karpathy: Globally? Maybe a few hundred people who truly understand the cutting edge. It's a small field, and the expertise is concentrated in a handful of teams. That's both a challenge and an opportunity — it means progress is bottlenecked by talent, but it also means that a single breakthrough from one team can advance the entire field.

SEGMENT 3: The Competitive Landscape

Dr. Sarah Chen: Andrej, who else is working in this space, and how does your approach differ?

Andrej Karpathy: There are several serious players. Some are taking the scaling laws and compute trajectory route, which is well-established but has limitations. Others are betting on reasoning models and chain-of-thought, which is newer and riskier but potentially more powerful. Our approach is different because we prioritize multimodal capabilities — most teams optimize for peak performance, but we optimize for reliability and cost at scale. That's what matters for real-world deployment.

Marcus Webb: Is this a winner-take-all market?

Andrej Karpathy: I don't think so. The market is large and diverse enough that different approaches will win in different segments. The real competition isn't between the companies in this space — it's between ai agents and the incumbent technology. We're all fighting to prove this is better than what exists today, and there's room for multiple winners once we do that.

Dr. Sarah Chen: What about the investment environment? Is there enough capital flowing in?

Andrej Karpathy: The capital is there — maybe too much, in some ways. The risk isn't lack of funding. It's that funding without focus leads to wasted effort. The companies that will succeed are the ones that can turn capital into concrete milestones — not more press releases, but real technical progress that moves the needle.

SEGMENT 4: What Comes Next

Marcus Webb: Andrej, what should people be watching for in the next twelve to eighteen months?

Andrej Karpathy: Three things. First, agent architectures — we'll see meaningful progress there, and it'll be measurable. Second, open vs closed source — there are developments coming that the public doesn't know about yet, and they'll change the conversation. Third, the policy and regulatory side — training efficiency is going to become a much bigger topic as the technology matures.

Dr. Sarah Chen: If you could give one piece of advice to someone listening who's trying to understand where AI is heading, what would it be?

Andrej Karpathy: Don't trust the extremes — neither the hype that says everything changes tomorrow, nor the skepticism that says nothing will ever work. The truth is in the middle, and it's moving faster than most people realize. The best thing you can do is engage with the actual data — read the papers, look at the numbers, talk to the practitioners. The signal is there if you know where to look.

Marcus Webb: Andrej Karpathy, Founder, Eureka Labs at Former Tesla AI Director. Thank you so much for joining us today.

Andrej Karpathy: Thank you both. This was a really enjoyable conversation.

Dr. Sarah Chen: And thanks to all of you for listening. This is TechNova — see you next time.