Quantum computing + AI + agents = AGI?

Mykola Maksymenko: A couple of months back, I came to a meeting with our team and said, “Guys, I think AGI is here.”

John Koetsier: Welcome. Give a 30-second introduction. Who are you?

Mykola Maksymenko: Hi, John. I’m Mykola Maksymenko. I’m co-founder and CTO of Haiqu. I’m a quantum physicist by training. I used to work in industry for a while, in technology management, and now I’m back to the roots, building the most performant quantum software for future quantum computing.

John Koetsier: So somehow you, as a quantum physicist, have been looking at genomics. How’d that happen?

Mykola Maksymenko: That’s actually one of the side projects for fun, let’s say. So I’ve seen recently interesting announcements from our friends, actually, from Oxford loading a genome on a quantum computer. So I was curious, like, okay, there is a genome, and you can actually download those genomes. They are open on the internet.

And I used our AI system that’s kind of an agentic operating system, which allows us to decompose a complex problem into simpler problems and then use our underlying operating system for quantum computing software to run something on the hardware. So I used the system with this input data, and I was surprisingly able to reproduce the results pretty quickly and was loading the genome and doing some basic operations, for example—

John Koetsier: I will take a break. So—

Mykola Maksymenko: —like making some small mutations to the genome and seeing how you can trace the mutations as you make them. So that’s not my specialty whatsoever.

John Koetsier: No, totally not. Totally understand. But it was pretty impressive, right? I mean, it was three years of work, and you managed to do it kind of in a couple of days, agentically and with your quantum computer. Is that correct?

Mykola Maksymenko: Yes. So that’s actually a pretty impressive direction where all the science is moving. Some of these complex research projects—in this case, it was Q4Bio. That’s a larger initiative from Wellcome, I think it’s a Wellcome Leap grant. And so they sponsored a bunch of teams—

John Koetsier: Mm-hmm. So, thank you.

Mykola Maksymenko: —to do some breakthrough research in different areas of life science. And in this case, just based on the announcement, even without looking into the papers much, the system was able to guess what was probably done there and what kind of algorithms can be used, and how we can encode that using some of the tools that our platform has, and in the end run it on the hardware.

John Koetsier: There’s a really interesting convergence happening here, right? Because we have LLMs and AI systems advancing in their domain. We have agentic systems, which build on those but are a little different as well, advancing. And we’ve got quantum computers, which are always advancing as well.

And you’re bringing all these three things together. That’s kind of fascinating and might lead to a speedup in a lot of different things, correct?

Mykola Maksymenko: Exactly. I would say it’s, in some sense, self-accelerating because a few years back, people were not able to run anything on quantum computers. Basically, it was very small toy examples. Today, we run things on hundreds of qubits, which are on par with, or slightly behind, what the biggest classical computers can do. And with AI help, we can now dramatically accelerate the discovery of these new algorithms.

And I believe that even with current quantum computers, we can actually already do interesting things in condensed matter physics, in chemistry, maybe in biology as well. And with these new agentic tools, we can now go over multiple use cases much, much, much faster. Instead of a three-year project, or instead of a half-year project for a PhD student, we can now quickly validate some hypotheses over the weekend.

John Koetsier: Wow. I mean, Sam Altman recently said we’re in the singularity, and I was kind of rolling my eyes at that. But you said these things are self-accelerating. I mean, that’s kind of perhaps what we imagine the singularity looks like. We imagine it a little broader than this. It’s pretty domain-specific, perhaps, but that’s a piece of it.

Mykola Maksymenko: Yeah, I would say a couple of months back, I came to a meeting with our team and said, “Guys, I think AGI is here.” So my discovery was, again, over the weekend, I was doing some projects, and it was my PhD work. It took me actually around half a year.

John Koetsier: Ouch.

Mykola Maksymenko: And agentic AI was able to reconstruct it and actually find the bug in one formula, which—

John Koetsier: Yeah, yeah.

Mykola Maksymenko: —I was overlooking. But the thing is that if you prompt it correctly, if you add some expertise and guidance, these things can be very powerful. And I would say it’s like a weak AGI in some sense. It’s probably not completely self-consistent, but you can already use it for many good tools.

John Koetsier: So this is interesting. It’s got huge potential for amazing, good things. It’s got some potential as well for misuse. If you can load genomes onto quantum computers and then just start making mutations, maybe you can invent some wonderful, amazing new antibiotics. Maybe you can invent some new virus. Maybe you can—the mind kind of boggles at the possibilities.

Mykola Maksymenko: Exactly. That’s already the problem I faced. So when you play with these systems and you suddenly go in the wrong direction, for example, with a genome, and you use Fable, of course you will not be able to do this. Or if you use these systems for some chemistry simulations, you also can get into guardrails very quickly.

So that becomes challenging because, on one hand, this is valid research. This is not something harmful. But on the other hand, some people can use it for completely harmful things.

John Koetsier: Mm-hmm.

Mykola Maksymenko: That becomes, I would say, a large trade-off in the industry right now. So who will get access to the most powerful frontier models, and who will not? Would some of the open-source models be as powerful or as guardrailed as some of the frontier models? Who knows?

So, yeah, we don’t know where we will get in one year from now.

John Koetsier: It’s Pandora’s box, and we’re opening it all over the world, and we’re in a race to open it because others are opening it. And, yeah, it’s an interesting, interesting world.

Where do you go next? I know you tried loading the COVID genome onto it as well. What else? What else are you looking at working on?

Mykola Maksymenko: Again, that was one of the side quests, let’s say, playing with this recent interesting announcement and trying now to load this kind of data. But we are currently working with a bunch of academic groups. So we gave our system to researchers actually working either in life science or in condensed matter physics to work on some of the hardest academic challenges in—

John Koetsier: Mm-hmm.

Mykola Maksymenko: —condensed matter physics, in material design, in chemistry, also in biology and genomics as well. One of the researchers actually is working in genomics. And the interesting quest is to showcase that, with the help of these tools and with the supervision of expert scientists, you can actually get incredibly interesting and new frontier results in a bunch of connected disciplines.

In our case, this is, of course, connected to quantum. So we provide the—

John Koetsier: Mm-hmm.

Mykola Maksymenko: —best execution layer to run this later on the quantum computer. And for us, the quest for Haiku and for myself in this case is to demonstrate that we can run interesting, large-scale applications of the frontier scale already today. So we don’t need to wait for 10 years from now for fault-tolerant quantum computers, that some interesting applications are in the vicinity.

John Koetsier: I want to double-click on that. And you hinted at it earlier, right? Is that we can do actual work right now. And we’ve had this feeling like, okay, quantum’s here, and there’s interesting things happening. I reported on one of them just literally yesterday: a D-Wave computer that AT&T is using, doing some interesting—solving in about 15 seconds what took classical computers about an hour, right? Which is network optimization for their network.

If you wait an hour, the network has changed. Needs have changed. You don’t have that problem anymore. But we always kind of felt like we need maybe 100 logical qubits, maybe a thousand logical qubits, to do really interesting, fascinating problems. And what you’ve discovered is, by converging AI and agentic systems and quantum, you’re doing interesting work right now.

How many qubits are you using right now? How many logical, physical qubits? What do you think the potential is right there, right now?

Mykola Maksymenko: Right. So where does all this logic come from? We’ve seen these so-called quantum supremacy experiments by a number of groups where, if you run something on hundreds of qubits, that can go beyond what’s possible to run on classical computers. We know that those experiments were very much artificial, in some sense, that they were very fine-tuned to a specific mathematical problem which has nothing to do with useful reality.

John Koetsier: Ha.

Mykola Maksymenko: And it made a big noise. But now the question is, okay, there is this narrow point where we indeed can do something bigger, better on the quantum computer than on the classical. Can we expand in the vicinity of that point? Can we go to slightly different problems, but maybe of similar complexity, which—

John Koetsier: Mm-hmm.

Mykola Maksymenko: —probably would have some usefulness? So that’s what we are trying to explore. And there are many such points.

It’s not each and every point, or each and every problem from the industry, that we can now solve on quantum computers, but we are trying to find either optimization problems or quantum machine-learning problems or some hybrid high-performance computing problems, for example, in diagonalization of large matrices. So there are a few cases where we know that we can actually squeeze interesting results from quantum computers.

John Koetsier: Mm-hmm.

Mykola Maksymenko: Now the challenge, the quest, I would say, for all this community is to find usefulness in that area around this supremacy regime. And that’s what people call quantum utility. And that’s where we are today. So it’s around hundreds of qubits to two hundreds of qubits, not necessarily logical. If you have hundreds of logical qubits, for sure you can do interesting—

John Koetsier: Yes.

Mykola Maksymenko: —things already today, but we are not there yet.

John Koetsier: Yeah.

Mykola Maksymenko: So-called physical or noisy qubits. But apparently there are interesting problems. For example, you can do simulations of molecular dynamics. One—

John Koetsier: Mm-hmm.

Mykola Maksymenko: —of the first demonstrations that we did with AI was demonstrating that we can use AI to create an algorithm to simulate neutron scattering on magnetic compounds. And that neutron scattering result was almost on par with what I did in academia, like 10 years ago, on the largest supercomputers.

It already becomes useful in some small aspects. For example, running these dynamical simulations of a small molecule on the quantum computer is just easier, more natural, more convenient than on the classical computer. It just takes more effort to map it onto the classical computer. It takes—

John Koetsier: Okay.

Mykola Maksymenko: —more time to run these simulations. It takes a lot of resources and heat to run these kinds of simulations. I would say quantum dynamics and quantum chemistry are some of the low-hanging fruits.

John Koetsier: What a wonderful thing. What serendipity. What an amazing thing, because we’ve been chasing this quantum advantage. We’ve been chasing this quantum supremacy for a decade or so, and the number of qubits is huge, and physical, logical qubits is huge as well.

And now, by adding some AI and agentic systems, you’re demonstrating quantum utility. Hey, useful—what we have right now. It’s kind of an MVP, minimum viable product, of quantum computing. Absolutely huge, because that provides a stepping stone to continue to grow, continue to grow, continue to grow, while you’re having actual good results today. Super, super interesting.

Did you know when you started a quantum computing company that you’d be getting into agentic AI to make it more useful?

Mykola Maksymenko: I think we had probably some intuition because the domain that we used was AI, Haiku AI. And that’s where we are right now. It started more from using AI for optimizing some of the low-level middleware flows and error mitigation, circuit compression, and compilation, while now the AI is on a different level. It’s more on the high level of orchestrating the algorithm and making it easier and more democratized for the broader community to build something.

John Koetsier: You said self-accelerating, and so you’re giving scientists and scientific teams access to this technology right now. They’re using agentic systems, they’re using LLMs as good as they can get, whether that’s Fable, that’s a lower system, whether it’s open source, open weight, whatever the case might be, and they’re using quantum computers.

I mean, how fast are we gonna get here? Just, I mean, talking real world. We’re not talking 18 months out when you’ve got more qubits or 24 months out. Just right now, you’ve talked about a speedup that you did, which is like, it’s gotta be a factor of 100x or even 1,000x, right? Because you redid the research that took you a year and a half in hours. You redid the research that Harvard did over three years in an evening or something like that. Talk a little bit more about the speedup in different domains.

Mykola Maksymenko: Yeah. In my personal case, I see that it allows me to speed up probably from half a year to a week. Just today, I talked with one of the partners who is using the system, and the feedback was similar. They did not arrive at a result yet, but they already have promising, very complex results, and they spent a couple of weeks instead of working for a year on those.

And they kind of did not have experience in that particular domain that they’re exploring right now. So—

John Koetsier: Okay. Yeah.

Mykola Maksymenko: —I would say this dramatically enhances how we are doing science and how we will be doing science a year from now. So we are focusing, in this case, on quantum computing. Some other people I know are using agentic AI for mathematical theorem proving. Some other people are doing some life science research and mining large-scale data.

All this becomes kind of—it’s very interesting how different domains can start to converge, because now you don’t need to be an expert, or a deep expert, in some domain to start working in this. With the help of these tools, you can advance faster and learn as you—

So these tools can actually even act as educators for you. They can teach you what to focus on, which papers to read, and you can advance much, much faster.

John Koetsier: It’s absolutely fascinating. I mean, decades ago, Steve Jobs talked about computers as a bicycle for the brain, right? Accelerating what we do. We’ve kind of laughed at movies like Iron Man, where Iron Man is talking to Jarvis and he’s inventing whole new materials in the course of a couple hours and building an amazing new tool or machine over the course of a couple days or something like that.

But we get some sense of that magic in what you’re talking about, right? We get some sense of that as we use agentic tools and AI tools. And the acceleration we’re gonna see in so many different domains is amazing and fascinating.

It also opens up the field for more people to do science, right? I mean, who has a grant to spend three years investigating X in some tiny little field of metamaterials or something like that? But, you know, if it takes you three weeks to get something interesting, which then you can parlay into some funding to go deeper or build or whatever, that’s huge.

Mykola Maksymenko: Yeah. So one of our friends who is using the platform, he’s saying that now you can put the agent over the weekend and explore various ideas, which probably were somewhere on the back burner, which you would never even try exploring, but now you can just do it almost for free.

And sometimes these small ideas can actually lead you to interesting results, which can then probably become even a major research avenue.

John Koetsier: Yeah, yeah, yeah. Wow. Cool. This has been fascinating. This is interesting. And this is a new development to me, at least, in quantum computers and AI and agentic systems. And I look forward to seeing how it plays out and what we create based on this.

Anything else to add? Anything that we’ve missed that we should chat about?

Mykola Maksymenko: I would say that probably a lot of these results would not be possible without our operating system, which is lower-level code, which actually we spent quite a lot of time making to optimize the workflows of quantum computers. And now what’s exciting is that agents basically learned how to use it in a very optimal way. So it’s kind of two—

John Koetsier: Mm-hmm.

Mykola Maksymenko: —layers. And sometimes you discover some usage which was not intended, but—

John Koetsier: Mm-hmm.

Mykola Maksymenko: —agents actually find that you can speed up some things or optimize some workflows with different kinds of smaller subroutines. In the past, I compared it to the age of classical computers in the 1940s and 1950s. You needed a lot of expertise to manipulate the valves and computers with switches.

John Koetsier: Yes.

Mykola Maksymenko: So that’s where we were, like, half a year ago with quantum computers. It was very expensive and very hard to run something with these machines. And now you can use an agentic layer to help you basically operate those lower-level complexities. And you can really focus on applications and interesting things which actually progress you—

John Koetsier: Yes.

Mykola Maksymenko: —as an industry, as a scientist, and progress all of us in terms of this quest in search of useful quantum applications, which we hope to get more of in the next year or so.

John Koetsier: Mykola, thank you so much for this time. Fascinating stuff, super interesting, and incredibly exciting. Thank you.

Mykola Maksymenko: Thank you. Thank you, John.