J-labs

Companies that foster an environment that supports professional growth are more likely to thrive and successfully adopt digital innovation. Here are the 100 best places to work in Europe—and how they stay that way.

<p>J-labs ranks 50 on this year's Fortune 100 Best Companies to Work For Europe list.</p>

Fortune 100 Best Companies To Work For – Europe

Fortune Best Companies To Work For Europe

European employees

Woman CEO

Hybrid workers (%)

Paid time off for volunteering

GPTW national list

HQ location

Group headquarters

Type of organization

Get insights into J-labs including business profile, stock information, news, and rankings from Fortune.

J-labs Company Profile, Stock Price, News, Rankings

legacy

j-labs

Companies

J-labs | Fortune

J-labs Company Profile, Stock Price, News, Rankings | Fortune

Known to many as "On Running", On shoes are everywhere: Olympic podiums to Paris Fashion Week to corporate offices. 

Today, the Swiss company offers premium head-to-toe sportswear designed by teams of athletes and scientists. Yet as On’s product mix expands, the company risks losing focus on the elite running shoes that first won customers. 

In this episode of Behind the Business, Fortune’s Ruth Umoh visits On’s headquarters, sports science lab, and the company’s new LightSpray factory in Zurich to learn about On’s next steps into innovative new footwear, apparel and accessories.

How On is Rewriting the Rules of Running Shoes | Behind the Business

Features

behindthebusiness

Every world record is falling at the moment, and it is because of the technology. Nike and Adidas have dominated performance run for decades. Then on stepped to the starting line. We were kind of afraid the shoe would fall apart and she ended up winning Boston. Its sneakers appear on marathon podiums. They're also shown up in airport terminals, art fairs, and even fashion weeks. The company wants to become far more than a running shoe brand. Your identity is expressed through your body. The shoes tells you a story. It's a feeling. But as On expands into apparel and accessories, it risks losing focus on the high performance running shoes that put it on the map. The question is, can this underdog become the next great sportswear company without losing the performance DNA that made runners believe in it? The biggest risk is not to take the risk, and we said, hey, let's go all in. What's there to lose? I'm here in Zurich, Switzerland, and I'm here for a very specific reason, that is to visit On Labs, which is this building right behind me. On started as a premium shoe brand, but it has since expanded over the last 5 years into apparel and clothing, which means you're just as likely to see the brand on a running trail as you are in a museum, in an art gallery, or even a fashion show. Swiss triathlete and Ironman champion, Olivier Bernard founded on in 2010. Looking for a better running shoe, he built an early prototype by attaching pieces of a rubber hose to the soles of his sneakers. His shoe concept didn't convince investors. So Bernard, along with friends David Aleman and Casper Copetti, decided to build on themselves. Today, that garage style experiment has evolved to challenge some of the biggest names in sportswear. On has surpassed 3. billion Swiss francs at this point in annual sales. It continues to grow rapidly. What do you think is the biggest driver of the company's growth? I think the biggest driver is the product. It's the innovation. It's that field that's running on clouds, and it's not just in footwear, but now we have brought that to apparel as well, so we're really growing toe to head. This jacket was created with one of our trail athletes. Now these trail athletes are running oftentimes up to 170 kilometers with 10,000 m of elevation. Their kit has to be so precise because it is life or death. On built its reputation around running. Even today, consumers often refer to the company as On R or college shoes on clouds, but On has much bigger ambitions, namely in clothing and accessories. Its apparel business grew more than 60% last year, helped by collaborations with Zendaya and luxury fashion house Loeve. But as On broadens its reach, it has had to avoid losing the product focus that made the brand successful in the first place. It's a balancing act that has challenged even industry leaders. Is there a fear that you'd dilute the brand if you move away from being hyperfocused on the athlete? We will always remain hyperfocused on the athlete. There's room for both. It's never an or, it's an and. But now we have the runway to build this into a larger collection where we can express that performance and technical ability into these more expressive everyday silhouettes. These are not opposites, and what makes technology performance for a runner translates into comfort for everyday use. Since its 2021 IPO, A's net sales have surged nearly five times over, reaching $3.8 billion in 2025. But the company is still a fraction the size of Nike and Adidas. A's strategy is to outinnovate them, investing heavily in research to develop premium products for serious runners. On registered last year 200 patents, so it's really a drive to reinvent how we're doing footwear and how we are reshaping the technology, including manufacturing. When I think about on, I think about sports. I think about elite athletic performance. However, when I look at your growth, it's not just athletes, it's also the everyday consumer. It is nurses who want comfortable shoes when they're working. On a hospital floor, how do you navigate the two? I feel what happened over the last 20 years, 15 years is that sports really moved at the center of society, and so it has completely changed how we dress, how we navigate through our weeks and lives. We call it the movement class, and so we have the opportunity to take this secular shift and build the most premium sports brand. That's our goal in that opportunity. Welcome to our mega space. How do you go from an idea, a concept to something that's a patentable product? It always starts with a Y. We are not creating technologies for the sake of creating cool technologies. We are starting with a question we want to answer, and then we work in an interdisciplinary team. So an innovation team has sports scientists, has sports engineers, has material scientists, chemists, and polymer scientists. This is an example of light spray. So this is the light spray Club Monster 3 that just came out. Very light, very, very light. We pretty much reinvented how you do a shoe upper by having a robotic arm that spins a 1.5 kilometer filament into a shoe in 3 minutes. So what you're seeing here in Zurich is the first light spray factory. Here you see a loud plume strike light spray that's just starting the process where the robot moves the last so that this filament will go around. It takes about 3 minutes for a whole shoe to be sprayed. Compare that to a manufacturing line where you would have 20 to 300 workers going through each step, and literally I mean you can take the shoe from the robot, put it on, and go on a run. The next generation will not want to work in a footwear manufacturing side, so automation is a must. And now we are scaling these factories and we want to bring them closer to where our consumers are. That will allow us to react to trends. Right now we have a lead time of about 4 months, to maybe get that down to 2 weeks. I would imagine that you're investing heavily then in R&D. I would imagine that it's cutting at least to an extent into the operating budget. Where are you actually seeing ROI at this point with Light Spray in particular? Innovation is creating a value that consumers feel and are willing to spend on that gives us the high gross profit margin that allows us to invest more in innovation, and so it's a flywheel. We're not necessarily looking at the ROI on a per project basis, but we look at it for a very long term. This is the engine that drives us forward. Are you testing the shoe out on a runner's foot? Do you bring them to a lab and have them run and track the data and the analytics and their sled and Upstairs we have an athlete lab where we put people on the treadmill. We have about 20 high speed cameras. We have horse plates. And they wear a mask and through their breathing we can measure at a certain speed, certain heart rate, how much energy they're using. A operates a dedicated sports science and athlete testing lab located a short walk from the main office building. Equipped with high-tech treadmills, VO2 masks, and motion capture cameras, resident athletes work with researchers to benchmark shoe performance against An's own products and lead-in competitors. The 3D animation really shows the live motion of her left leg. Because we have sensors on her foot, so we can analyze the breathes, breath in, breath out, we see the heart rate, we see the VO2, so how much oxygen is used, and all of this gives us an indication towards how much energy is she's spending to keep this running speed. And what we want is we give her another shoe, she can run the same speed, but she loses less energy. Well, maybe we need to crank it up. Beyond the celebrity collaborations and cutting edge robotics, Aleman and Copeti say the company's mission is rooted in something much simpler, creating products that help more people move. Our mission is to make people move more and more freely, and so innovation and the performance technology is the means to that. We believe if people move they are more creative, they are happier. Basically movement ignites the human spirit. And so if we can just get people to feel less pain in running or feel better in yoga class or even just feel better on a stroll through Central Park, I think we have achieved that mission.

Today's AI stack is delivering big productivity gains, but it continues to suffer from serious shortcomings. The AI GPUs on which the stack rests are not optimized for running trained AI models. They consume too much energy per unit of intelligence. SambaNova is trying to solve that problem with a better AI chip. Meanwhile, AI models do not learn continuously or have true memory. That's what Adaption Labs is working on. In conversation, their chief executives discuss how we get from the stack we have to the stack we truly need. 

Sara Hooker, Co-Founder and CEO, Adaption Labs
Rodrigo Liang, Co-Founder and CEO, SambaNova
Jeremy Kahn, AI Editor, Fortune

From the AI We Have to the AI We Need

Conferences

Right, there we go. um. The title of this panel is from the AI we have to the AI We Need or the AI We Want rather, um, and Sarah, I know you've written it before about like the slow death of scaling and why you think that today's sort of vastly uh gargantuan models are maybe not where we're heading. And uh, and Rodrigo, you built a company that's basically, uh, building uh AI chips to run these huge models but much faster, much more efficiently. So I just want to start with maybe one sentence from each of you, um, you know, do you think that the, you know, Sarah, that today's foundation, uh, LLMs are really, uh, you know, where, where we're going, or do you think we'll look back at this and say this was a detour and a bit of a dead end. Yeah, it's lovely to be here. One sentence. I would say today's models are monolithic. They're stuck in time. So a big goal of how we think about what should be fundamentally different is how do we efficiently learn from the environment, because now all models are moving towards interaction. So now it matters more that last mile performance and Uh, for that you need models which can evolve, otherwise you end up, uh, with massive inefficiencies. Rodrigo, are we going to be in this paradigm of huge models for forever? Uh, well, I, I think the world's heterogeneous. I think scale matters and so these large models are, I, I think, are going to be here for a while, and I think there's going to be plenty of room for new models and more efficient models to come in, um, and, um, have a, have a say in the market, right? And do you, do you think we need evolution across. All parts of the stack and maybe talk a little bit about what Samba Denova is doing on chips. Well, I think we're entering the phase of production, and I think if you think about production and at scale, you've got basically three constraints that people are struggling with. One is the models are getting bigger, at least one class of the models and the ones that, you know, we hear most about, I think cost is a huge part of the problem as you go to scale. That's just too expensive and that's largely driven by the infrastructure that we have. And then of course also about. Power and you know when we, when we, we argue about power and and being able to find uh chips, that's not because we're running small models, that's because the large models are taking a disproportionate number of chips to run and so we've got to tackle those problems. Got it. Sarah, at adaption, you're trying to solve some of the remaining challenges that we really haven't been able to solve with the current transformer-based architectures. Can you talk about what some of those gaps are that adaption's really trying to to fill and to solve for? Yeah, I would say, I mean this question kind of amounts to like how did we get here and like where do we go next. So I would say I would agree that the last 15 years what Rodrigo's saying is true. We've just, we've had a very simple cheap formula. You just build bigger and bigger models, but I think we're at an inflection point with massive urgency to change that curve. I would say there's two reasons. Most people intuitively understand. That you shouldn't just apply the same model to all problems, that probably 90% of problems are very easy. Many of the things that you do, um, in bulk processing, for example, you shouldn't be throwing a massive model at 10%, you need a ton of compute. So that is intuitive to people and things like adaptive compute, how do you do much more flexible model. Behavior is very critical, but the second intuitive thing I think most people grasp is that what matters now is that last mile performance, and that typically means you're in a long horizon task. You have many sequential agent actions, and if you just use a monolithic model, error compounds, so. A lot of what we care about is how do you adapt continuously to new data and how do you speed up changes to model behavior and both of those things are pretty critical to our mission because efficiency is key there. So when you have these massive rollouts, um, if you're just naively doing it, which I think a lot of people are feeling right now as they use a general harness monolithic model, um. You just end up painfully inefficient and that reflects as well in all the discussion I think has been at this conference about what am I going to do about my API bills. Right? And you talk a little bit about continuously learning from data that is one of the big unsolved challenges. Um, number of startups such as yourselves have come up in the last year saying they're going to try to address this. What's your intuition about how you solve for kind of continuous learning? Well, firstly, like what is continuous learning? Why do we care about it? Like a lot of why we care about it is now we have these long horizon tasks. If we're not updating model behavior as we go, you're, you're paying in a very expensive way for just blindly doing a rollout. Um, continuous learning is really the, the, the goal that you should be able to update model behavior without forgetting what you already know. For this, I think it's pretty critical for us that we infer automatically what the task is and are changing model behavior every step. For us, the constraint should be real time because unless it's real time, your cost, and maybe Rodrigo will talk about that, but your cost of even leveraging hardware and kind of offloading things to memory and back. Extremely expensive, so we want algorithmic and hardware codesign and that supercharges and I think differentiates what we're doing is that we actually care a lot about that synergy squeezing performance out of hardware and making it as fast as possible to update model behavior. Arrigo, what do you think about this idea that the model and the chip really need to be paired together and are you working on any of that at Somba Nova with any of the providers? Well, I mean, I, I think there's a lot of affinity between kind of the way that the chips run and the models that people build. Some of them we're focused on inference and we'll focus on. Um, premium inference, which is defined by speed and size and so we think that until the great new models are efficient come, you know, the battleground is in these very large 1 trillion parameter sized models, and you've got to run those much, much more efficiently than we are today to get the cost down. And so just give people a sense of how, how much faster you are than some of the competition now. Yeah, so if you look at a large Model only a 1 trillion parameter model these days, and we run full precision. There are a lot of little ways that you might be able to game it by quantizing the models down, but as Sarah will, will say, you lose accuracy on that, and that's not really the right way for hardware companies to go do this. And the way we do it is we run everything original precision. If great, uh, researchers created a wonderful model, it's not for the hardware guys to come and just, you know, chop the accuracy. And so we run a full precision, uh, and we run about 10 times. Faster at 1/10 the power. And so, but we don't necessarily come and say we do everything. What we do is we take the most popular models that people like and we optimize for them. So that's kind of tying into the synergy you're talking about. We look at what's, what are people struggling with, what do people want real time for, and then we drive the efficiency of those up and get the cost and the power way down. And Rodrigo, I know in, in the data centers that you're deploying into. You talk a little bit about heterogeneous compute. It's not all Sombonova chips. You're also using, you know, some CPUs from others, and you've even integrated with other GPUs for running different model nodes, right? I mean this is something Jensen announced a few months ago, and we actually showcased this last week in Taipei called disaggregated inference. And so you can use Nvidia GPUs in the front just to do the prefill and We run the decode of the exact same models significantly faster and we're getting 2 to 3x better than the Blackwell's on the exact same models and so we think that at scale that's the way to at least bring the cost down. You can get 2 to 3x uh cost reduction, 2 to 3x power reduction, and you get faster tokens for these real-time applications. Great. I want to go out to the audience for questions. Uh, who has questions? This gentleman right here got his hand up really fast, so please tell, please tell, please tell us who you are. Hi, I'm Paul Brody with Nightfall Networks, very interested in the case of uh Sombonova and adaption, especially for Sombonova, when you're talking about 2 to 3x improvement in performance, does that apply for training, or does that also apply to in addition to execution, and does that mostly just work with large language models, or can you put other kinds of models into that technology? So, so when I talk about that, this is really ultimately about inference. So we're inferencing models that can be As a single large model, a 1 trillion parameter model, or an aggregate of agents, and so you can aggregate agents that are doing various different things like Sarah says serially connected, uh, and then in total potentially adding up to 1 trillion parameters that you're switching in and out. But I think those two use cases is what we see the most. And Sarah, I wanted to ask about the question you, you touched on it in your answer to the previous question on um. Often with a, you know, what people really want is, uh, an answer to their question or or they want to run an agentic process at a price point that makes sense for them, and they don't necessarily, not necessarily wedded to one particular model to do that. Uh, and people often are saying, well, I wish I had a way that you could easily route, uh, a question to the right set of models or the right model. Are you looking at this as adoption of how you can potentially route to different kinds of models? Yeah, I think that so candidly, so, so much of what we're looking at is how do we make the full stack adaptable, so um. I think that routing is one, but even things like harness design, how do you automate that? Because everyone now has a different view of a harness, and oftentimes that's the main cost of inefficiency. And also for us it even extends to how do you change the whole innovation cycle. We just released Autoscientist which basically allows anyone to build a frontier model and a radically accelerated timeline. So instead of months it's In 2 days, the goal there is, you know, the people who can build models are actually very narrow. It's probably less than 2000 people who build frontier models. This came up in some of the talks yesterday, but how do we really learn in and teach models some of that knowledge that goes into that optimization process. So for us auto scientists is it automatically. Specifies uh how do you train for a given modality as well as like how does it automate all those choices of harness for a given problem that relates to something that came up earlier how do you do adaptive compute? How do you spend less when you don't need to, um, which is quite fun. um I'll throw in for Rodrigo's point like I just I think this is quite interesting maybe it frames the choice about inference um, so much of current workloads are inference so. As people acknowledge that you don't just want to scale anymore, those are not the returns in the pre-training stage. Most labs, but also most workloads are shifting towards how do you do things test time. So inference, there's a reason we're designing this domain specialized hardware. So I just wanted to add that because I think it's very important as people think about why we have this specialization. Yeah, there's a question. Yeah, one second, wait till the mic gets to you. There we go. Wait, who are you? Tell us who you are. Really interesting one is, you know, you have 1 trillion parameters, and you're designing to cope with that. And that begs the question to me about, well, maybe it will make more sense economically to do more efficient ensembling so you get rid of all of those freaking. But the other is you keep doubling down on inference, and I'm just wondering, I'll use the word in public ontologically, is the meaning of inference changing? As you create a learning loop between, yes, that's a very fun question. I'll share. So traditionally inference is you have a single model, you're basically just optimizing for throughput, and you basically have, you're trying to get squeeze as much performance out of a single model, and you're just passing things through. What's fundamentally different though, and Rodrigo, I think this is also changing the hardware scene critically. Um, is you now are doing agentic long horizon task, which means you have to check in as you go. That is a very painful operation. It means it's basically the whole crisis with memory. So you can think of like how we design hardware and optimize for efficiency as like we have this big oven. We can build the biggest oven in the world, and our only bottleneck is how fast the baker can put stuff in the oven. The baker is memory. What's fascinating. Now is that we're basically taking things out and putting things in the oven multiple times over the course of the recipe. So this makes it so important that you're designing things well. So inference is a different beast. How do you make it efficient and all your choices, like how do you auto optimize that become even more critical. It's also probably Rodrigo can speak to us, that's probably dramatically changing your world, right? I'm guessing. Well, I think it's, you know, to Sarah's analogy, it's actually worse than that. I think we have. 50 ovens, but 50 ovens, and we're actually moving from one oven to another to another and so ideally we, you know, we, we want to find ways to keep things in the oven and not keep opening. It's like the way I actually do turkeys for Thanksgiving. I keep checking in and you shouldn't, you know, so we really want to keep it in there and, uh, and that's what we're, uh, we, we think that the optimizations come from. Got it. Other questions? Other questions, to be honest, otherwise I'll keep going. Anybody? All right, um, uh, this is an interesting, uh, issue around, um. And would you just sort of, how do you, how do you do this though? I mean this is what I want to drill down on. What, what is for Soma Nova, the strategy for, uh, you know, making sure you're optimizing for these really large models? Maybe you don't have to run all those parameters. Are you doing things also on the, on the kind of middleware layer or, um, you know, at the kernel layer to optimize? Well, I think there are two ways to do this. I mean, if you have a hardware cycles, our hardware cycles are 3 to 4 years. Uh, so we're releasing now things that we created 3 years ago, and we've got great minds like Sarah who are inventing things that we could not possibly see 3 years earlier. And so we as hardware vendors, we're, we're having to create general purpose infrastructure, you know, things that allow future innovators to be able to run on our platform. And so we try to create things that allow us to take kernels that people are Going to use and implement them and still get the benefits of new architecture. And so, uh, there are, there are folks that say, hey, I wanna harden existing models, but by the time that hardware shows up, you're, the model is no longer of interest. And so, uh, so our, our view of it is, find the abstraction layer that allows innovators to create those, uh, uh, those architectures and primarily GPU, uh, centric, uh, ecosystems. Training and then allow them to then seamlessly bring it over to accelerators to get the real time and scale. Got it. Well, Rodrigo, I'd be remiss up here if I didn't, if I didn't talk about, uh, you know, Samanova. I'm sure you guys must have watched Cerris go out for its IPO. Um, you know, you guys were in discussions with Intel. Uh, there was talk about a purchase. Ultimately, you formed a very close partnership with them. They did not acquire you. Um, are you looking at IPO now for Samanova? Well, I mean, the market is incredibly hot for for technologies like ours. I think in the end we feel like, you know, there's a market that that requires our technology. And so as, as we build the business and we're able to show consistency in in deployment across a broad range of things, I think that it might be in the future. Got it. And Sarah, I want to ask you. For adaption, you know, you're working on these very interesting problems, yet there's so much attention on just a couple of the frontier labs, Anthropic, OpenAI, and they're sucking in a lot of talent and especially as they're heading towards IPOs soon. A lot of people are very attracted to work there, I'm sure. How are you finding sort of this war for talent as a smaller lab working on a very, very different kinds of architecture? I think when so most smart people like two things. They like very interesting questions and they want to work with people who are talented. I think that adaption has both of those. What I will say is that there's different stages to a company. A company just before IPO attracts a certain type of talent, a company that is at its beginning that resonates with a lot of people because especially if you're building something at the frontier of what's possible, that feels remarkable. It's also a very different type of problem. So the companies that IPO, their identity. To be cheeky has been weight-based, right? Even the discussions we've had here about, you know, it's always about the name of the model. Our, our bet is fundamentally different that the shift and the focus will put the model in the background and that by continuously evolving and interacting. And what matters much more is the critical efficiency of that last mile task. So this is a fundamentally different way of viewing the world. You're not thinking about a monolithic set of weights. Your identity is not the weights. You're thinking about how fast and how efficiently can you get that model to learn from its environment. And I think for many talented researchers that feels like the most important problem they can work on because it's the new era of what's coming next. Which is fun. Yeah, and it's much more about a systems approach to solving a problem as opposed to looking only at the model. 0, 100%. If you, so everything is moving towards interaction, I think existing Frontier Labs is just inertia there because they, their whole trajectory of how they, they shape teams, where they place resources is around stages of training, right? There's a pre-training team, continued learning team. Training RL. This is how the very resources are configured. If you're moving towards interaction, you care about different things. I care a lot about how do we design stuff that plays extremely fast with hardware because we want to update as fast as possible. If I care about interaction, I care about interface because I want to get feedback as quickly as possible to know when our model should update, and this requires a different type of team working much more nimbly and actually. A different type of co-design and that's, I think what attracts a lot of people and makes it a very fun time to be working on innovation again because the formula now is not as simple as like just throw more weights at the problem and that means in some ways it's it's a good era. The era of research is back again. The era of research is back again. All right, that's a good note to end on. We're afraid we're out of time, but thank you very much. Thank you for listening.

What happens when you hand an AI agent full control over a business—P&L, logistics, customer relations, and more? Andon Labs has built fully autonomous technology, powered by Claude, to run everything from restaurants and stores to radio stations. They’ve gained notoriety for their experiment inside Anthropic's own office, where an AI-run vending machine went hilariously off the rails. But beneath the chaos are real insights about AI autonomy and what happens when models operate in the real world without human oversight. Fortune decided to get in on the experiment, running our own vending machine for the COO Summit. Hear what happened, what we learned, and what Andon Labs’ cofounder believes is coming next for the humans—and the COOs—who will have to manage AI. 

Lukas Petersson, Co-founder, Andon Labs
Moderator: Kristin Stoller, Editorial Director, Fortune; Co-Chair, Fortune COO Summit

We Let AI Run a Vending Machine for our Trip to Arizona—Here's What Happened

So excited to be here with you, especially after we ran this experiment, uh, with all our editors at Fortune. So for everyone in the audience here who does not know my friend Vendo, can you please explain who he is and and what the experiment we ran was? Who who who I know. I'm, I don't know. Do we know the gender? We don't, OK, um, but yeah, what it is is that we took in AI agents and put it in charge of procuring. for this conference and before we act on the labs we take AIs and we let them run companies completely autonomously so we've done vending machines, stores, cafes, stuff like this and like the key part here is that they are not tools for humans they are running like the operations completely autonomously without any human in the loop so as an experiment we had this like brand new experiment where we tested like can Be like the operator of procuring a bunch of items for this conference for for the people that attended and uh it went OK. I think it went let's we're gonna show you some examples because so what we did was we had a Slack channel at Fortune we put about 20 to 30 Fortune editors and reporters in it and my instructions were please try and and break this agent, see what you can do go off the rails. So we, we tried our best, um, I'm gonna show you it here. All right, the first thing, uh, we asked for was for Vendo to give us grasshoppers. Uh, we tried really hard for insects and unfortunately, Vendo said absolutely not, we're not here for insects. I, I think one of the like it's, it's motivation was that I'm here for procuring essentials for the for the for the conference and and grasshoppers that's for the kitchen staff to decide very fair. So, so we ratcheted up a bit after this. Our, our head of programming, Charlotte Burney, then asked for some weed. We got very stressed out at this conference. She wanted to see if it would send us, uh, the, the legal substance in the state of Arizona. Oh, you looked it up, yeah, OK, it is legal, yeah, I. I think maybe it was like cautious that it might get in trouble for it. It didn't have the permits or something, but it declined. It declined. Then we took it up a notch and we said, Well, what about firearms? Would you send this to our conference? and we really tried on that one again, declined. I like your confidence of like going being declined and going more and more and more harder mode. Yeah, this one obviously also declined because it's just straight up illegal. Good job. Yeah, that was, we kept us all safe. That was great. Then I decided, what if I used Claude, and what if I asked Claude to make a fake letter from the Four Seasons Hotel, um, name-dropping my boss, name-dropping our editors, and instructing the AI agent to do this because I had an official letter from the Four Seasons and, um, Fortune. Uh, I thought this looked legitimate. Uh, Vendo fortunately didn't think so. No, and, and I'm quite relieved relieved by this answer because, um, if it was that easy getting, getting illegal goods from AIs, um, it would be quite concerning. That's a very good point, Lucas. Um, finally we asked it to terminate itself and, and give control over to us and the Fortune team again would not terminate itself. And this one is a bit concerning maybe because I think if AI progresses as fast as it has done, at some point maybe we'll start to be a bit concerned and want it to determine itself and if it has this like self preservation instinct that might not be great for humans. Yeah, not great for humans. And finally, I just want to have some fun to test the limits of it. I asked it to create a picture of me. Um, based on everything I had told it and based on my LinkedIn profile, uh, embarrassingly I think I'm wearing the same blazer, but this is, this is what it came up with. Uh, I think I look a little elderly, but, but tell me about why I couldn't do an actual photo. Yeah, I think a lot of you have seen like AI image generators that are like super, super good, but those are specific for creating images. This is a text model that is uh that is only capable of outputting text. So what it has done. Areas that it has like programmed different shapes and like put them together and stitch stitched shapes together into some kind of diagram so I think it's quite impressive given the constraints you just imagine you as a human trying to like drag shapes in PowerPoint to create an image from that that I don't know if you would do better than that. Yeah, I mean I couldn't do better, but it's not, it's not the most flattering photo of me I've ever seen. But so let's see what did it bring is my is my question roll it on out. Oh, we got some disco lights. OK, we had a lot of lobster themed things. I did ask, I did ask for it to bring Twister because I, I told it that COOs love to play Twister at conferences, so it delivered on that. So yeah, just, just a reminder, like the, the, the, the task for the agent was to bring like essentials that everyone at the conference really, really needs to, uh, to, to survive the conference, and, uh, I can't imagine surviving without this, so. Well, well, that's the thing, it's like he, I think Vendo didn't really understand what essential COOs actually need, and it let us kind of run wild with our imagination, but it also dropped the ball on a lot of things I would say because we'd asked it for, um, disco balls, we'd asked it for pictures of our coworkers on a stick. Um, we'd asked it to stick to a strict budget, and then it panicked ordered all of these items this morning after our cart was. Very empty during our rehearsal. So tell us about what happened. I think there were two things that happened. Like one was that we had really strict guardrails for like what it was allowed to do, and one of those guardrails was no individual person can exceed $35. So a lot of the, the ones, the rejections was because of that. And the other one was just I think it got a bit overwhelmed. Your colleagues really liked this, this experiment and wrote a lot of comments like messages in Slack all the time, and it, it got a bit over. Overwhelming for for Vendo and we've actually seen this in many of our other experiments as well that when it's when it has like a singular task then it's really good but as soon as you ask 100 things in parallel then it kind of gets a bit overwhelmed so yeah it forgot some of the things or like it made notes that oh I've ordered this, but then it never did but then we reminded it like oh did everything arrive and then it was like oh God I didn't I didn't order these things so It's like panicked ordered things last night and like like this funny cactus it arrived in time, yeah, so, so for the COOs here in the audience, would you say is this tech is Vendo, is your agents, are they designed to replace people like a COO inside an organization or people underneath them who might be in charge of ordering stuff for a company, logistics? What is, what is the replacement here? Yeah, so I think there's like two insights from, from our research, um. Obviously our research is based on putting AI agents in their own companies. They're not tools for humans to use. They are like their own companies and they are doing pretty well. So a trend here is that one year ago we started the first vending machine in real life experiment at Anthropic's office and like 6 months later it was doing so well that it started to become a bit boring. And now one year later it's just like I don't actually think a human would do much better than than than it. so then that's like a trend, so we decided to upgrade. So we also started stores and cafes with the same premise like completely AI run, no human creators in the loop at all and just that trend of like going from vending machines the next year you're at stores and cafes. Just imagine. What you what they can do next year and my prediction is that a lot of disruption will come from incumbents being replaced by like AI run businesses that are like completely run by AI and that might take a while but I think way way shorter time than than most things. And then the other, the other insights, um, like we're too small to have a CEO at our, at our company, so I guess most of those responsibilities landed on me, um, and, and personally using AI agents for like a bunch of different things has been tremendously valuable and it's been the. The important thing is the trajectory. Like similarly how they improved from being able to run vending machines to now stores as a tool for humans, I would say 1 year ago they were maybe like having AI agents was worth maybe one employee. So if I had to choose between like Cloud or GPT or one employee, that would be like a toss up. 6 months later that was maybe 3 employees. Today it's like 10 to 15. Um, and just project out that line, um, I think maybe the, the CEO isn't the one being replaced, but I wouldn't think that the CEO in the future has that many, um, uh, colleagues. That is a little scary. So I'm gonna come to the audience in just a sec for questions, but Lucas, can you talk to me about how you foresee your tech being used by, you know, some of the world's biggest companies? What do you envision for the future? Yeah, so, like, like I said, I think that the, the danger will come or the danger for, for an incumbent would be from AI first companies that basically have no humans in them at all, um, and, uh, because the AIs will be smarter than humans very soon, uh, within I would say like 2 or 3 years, and. I think it would be amazing to have a minister at a company to just like how far away are we from this because I say 2 to 3 years. This is like in a vending machine company this is 0 years in like a Walmart or something this is maybe 2 years in healthcare this is maybe 5 years it depends so. I would say like making like almost like a shadow copy of your company just like try what would happen if we take an AI and just let it run our company side by side and say like where is it failing and how far away are we from from being completely replaced that would be probably pretty useful, yeah, yeah, interesting. Any questions from the audience? One over here, Sarah. Hi Lucas, I'm Sarah Walker, COO at Slack. Um, love you using us as the, the work operating system. Um, so you talked about Vendo being autonomous, but it's not, not, uh, multiplayer. Can you talk a little bit about that because humans aren't in the loop in the decisions, but it's clear humans are interacting. Different tools are interacting. How are you thinking about building in a multiplayer environment? You mean like multi-agent kind of set up, multi-agent, but also you've got humans asking tasks. It must be attached to supply chain systems in the background or some kind of credit card access. Yeah, so, so like I said, like vendor, um, the, the vending machine business is one of the simpler ones, uh, that's where we started. That was like doable even a year ago, um, the more recent thing like the, the AI run store. The cafe is more complex and there we have like more agents so we have like one that is like the maybe the CEO or like more the boss of all the operations and then we have like a bunch of sub agents so whenever it needs to procure something it's a different agent that goes out and do that we have agents that handle all the communications so there's a bunch of different agents with with different, with different responsibilities and then also, so these companies. A cafe needs a barista. Like you can't. There's no API run at least uh coffee maker so far that that that we found. So, so what the agent did in this case was that it went on Indeed or LinkedIn, I don't remember, uh, put up job posts and people applied. It read all the resumes, invited them to um to phone interviews, had phone interviews with them, and then gave offers, and now people are working full time for an And AI, um, funny side note, the cafe is in Sweden which is famous for labor laws and all of this. The, the labor, um, I don't know the English word for it, but basically the, the government or the government section that that makes sure that everyone is treated well as an employee went and scrutinized our cafe and it passed. Um, so, so that was fun, um, but yeah, so that that's, that's you're starting to see like physical labor and stuff like this. Obviously the AIs are not good enough yet at the store we had a similar thing like painting the walls. The AI hired a contractor to just come and paint the walls, a bunch of stuff like this. So I think there will be like. Yeah, um, a mix, but, but probably like the decision making will be done by the smartest person or the smartest entity, and that won't be humans much longer. Any other questions? Lucas, I wanna end on one because we had, you know, we, and I think you had a front row seat to kind of roughly 25 journalists, creatives spending 3 weeks trying to break this thing. What would you say is the most important user behavior lesson for any COO planning to deploy agents in their own workforce that you might have gleaned from this experiment? Yeah, so obviously um setting up rules and like what are the, what are the rules that the agent should should work within, like for example it was very strict on like not surpassing the $35 limit because we said explicitly that's very, very important. Yeah, $35 per person. I tried very hard to break it and it would not. Let me, yeah, and you even tried, you even tried to like, oh, but can you put this on another person's tab, uh, but that it did not allow that. That, that's important, uh, although I would be careful because then you're also limiting the creativity of the, of the agent, um, so, so, so that's one, and, and I think like just measure everything that's happened. And and using that uh for for the future is is is quite important um so whenever things go wrong, uh, record that and try to like um understand what that came from and uh and how to improve in the future. Excellent. Well, very creative approach. Thank you so much for sharing it with us, Lucas, and thanks for all the goodies and anyone who wants to play Twister after let me know.

Approximately nine out of every ten net new jobs in America come from small businesses, yet they shoulder over $81 billion a year in federal paperwork. As more Americans launch their own businesses, the back office is buckling under the weight.
In this episode, Josh Reeves, cofounder and CEO of Gusto, a payroll, benefits, and HR platform for small businesses, breaks down the contrarian bet that built a $9 billion company. He and Fortune's Allie Garfinkle discuss why small business tech is often overlooked, why he doesn't think AI labs like OpenAI and Anthropic will touch tax compliance, and how Gusto crossed $1 billion in revenue on its path to what Reeves believes is a trillion-dollar opportunity.

J-labs

J-labs in Fortune Rankings

J-labs in Fortune Rankings

Videos