Building AI Workers for the Physical Economy: Robin Van Aeken, CEO of Magentic, on Agents, Manufacturing and Supply Chains

0:00This episode is sponsored by Cpoint, the business account built for startups. If you're a startup founder fed up with finance admin, you need Cpoint. Cpoint is different to neobanks because it connects all your bank accounts and your Stripe account to see your cash position, burn and runway instantly. It automates bookkeeping by pulling invoices and receipts from your whole team's inboxes, so you just sync everything to zero and pay outstanding bills with a click. It has a 3.49 % yield on treasury and real human customer support. Find out more at seapoint.co. That's S-E-A-P-O-I-N-T dot co. Use code UNICORN for a free month.

0:35Seapoint Treasury is a money market fund. Rate recorded at 1st of October 2026. Rates are variable and subject to change. Capital at risk. Hello and welcome to another episode of Rider Unicorns. Today we're joined by Robin van Aiken, the visionary founder of Magentic. Magentic is revolutionising legacy manufacturing with AI digital workers or majors. So we're interested to find out what they are and how they work, but ultimately they identify and deliver significant saving opportunities. So Robin, welcome to the show. It would be great if you could tell us a little bit more about these majors and how they work and how companies can set them up and the type of impact they have on the business.

1:18And firstly, thank you so much for having me on the show. Delighted to be here. I'm Robin. and co-founder and CEO of Magentic. And at Magentic, we build AI digital workers or majors, and they support some of the biggest manufacturers in the world, starting in the procurement and supply chain space. And to really understand the purpose of our AI agents, I think it's worth framing up a little bit the world of manufacturing companies today. What we've really seen, and it's happened in a really big way is AI selling digital products to other digitized companies. So for example, the first and really biggest moving categories within AI were AI writing software, AI training, AI applications for other AI companies.

2:01And that really got us from zero AI revenue to $200 billion of AI revenue. What that's come at the same time with is the physical world is really starting to feel the effects. So we've got the biggest CapEx cycle in history with something like$7 trillion of new CapEx going out into every single layer of the AI cake. So how do we get the power, the data centers, the chips, the cooling, the water systems, the conductors, the cabling, all of those components to really make AI work are all coming together. And so what we're seeing is this huge wall of demand hitting a supply constrained industry where it can take years to set up procurement and supply chain processes.

2:39Sometimes you're dealing with tens of thousands of suppliers making very intricate pieces all over the world. And doing this with the same team we've always had in a world where complexity is skyrocketing from the demand side, but also some supply side shocks we'll talk about later, is increasingly problematic. So what we do at Magentic is we bring a set of digital workers. They function a lot like employees within your normal procurement supply chain teams, and they join your team and they just start working. I'll give you a few really concrete examples. So we're lucky enough to work with some really incredible manufacturers, companies like Siemens or Coca-Cola Europe Pacific Partners.

3:14And what we do with them is day one, our AI agents or majors, they will join their systems, start to read through all of the spend, the orders that they've got in their ERP systems, and then find ways that we can make things more efficient, cheaper, faster. So for example, we might need to order a new kind of semiconductor from a supplier in Spain, but actually there's a German supplier who can deliver it three days faster. And Robin, is this work that is currently being done by humans, or is it work that the humans haven't got around to doing because they're under work with other things? I think that's why procurement especially is such an interesting place to play.

3:54I think a lot of AI goes after efficiency type work, which is seeing there's this human work, we can do it three times as quickly, we don't need the same number of humans, or those humans can do other work. I think there's such an interesting opportunity within procurement to not just do that, but to also tackle the fundamental spend that companies have. So for example, take some of our big companies, it's very common, they might have$10 billion of procurement spend per year, they might have a team of 1000 people managing that spend. If an AI company narrowly went down after trying to achieve people savings and managed to reduce the size of that team by 50%, that would be 500 jobs if you pay each person 100k.

4:32That's something like$50 million per year of total savings. That's substantial. But if we can deliver a one to 2 % savings on the$10 billion of a year that they're paying to their suppliers for raw materials, for parts, that's 100 to 200 million. So the numbers in the procurement supply chain space where we play is much more about making the people in the teams far more effective at managing that spend, rather than going after the efficiency type savings you might find there. Yeah, I mean, it's incredible the impact that you can have. And you talk a lot about human in the loop, and we're all hearing human in the loop, and this is sort of how AI is being rolled out to enterprise.

5:09But can you talk about how human in the loop allows you to embed yourself in these huge corporations who require a level of trust, and how the humans are playing into that feedback loop, and whether you're using their input to improve the product that you're actually building and forming a data moat over the long term? So we spend a lot of time thinking about what is our moat? Where is AI really winning? Where is it having the biggest impact. I started with the example of AI writing software is one of the fastest moving areas. And if we really dive into why that that is, our theory is that it's because of the speed of feedback loops you get in software writing, where an AI, a coding agent can change a line of code, run the piece of software, see if the changes worked.

5:52And then if it hasn't, try again and again, again, and can do this thousands of times a minute. In the world of procurement supply chains, this can take much, much longer. If you think of an example for like doing an RFP, a request for proposal, we'll go out, we'll find some suppliers, we'll email them, we'll ask them if they could deliver something to us. They'll email us back a few days later, we'll scan it all, we'll compare it. And what that means is that when you're trying to get AI to superhuman levels of capability in the domain of procurement supply chains, where we're focused, the fastest way isn't to try to tackle it in the old fashioned way of working in the system that they have today, but to artificially create these feedback loops where we can capture the incredible knowledge that people who've been running became a supply chain team for 20, 30, 40 years have built up during that time of, hey, this supplier over there, I know they've got this access and those parts will come in those bits.

6:41And if we're doing new product innovation, then this kind of supplier will work best with us. Capturing that knowledge through human in the loop processes allows our AI agents to learn at the pace of coding agents rather than the slower world that manufacturing typically has. Yeah, and Robin, one thing I noticed when you introduced the company was that you didn't say we're AI for procurement and manufacturing. You presented the fact that the world is changing. There's this huge pressure on like physical infrastructure. And that all leads back to manufacturing components and etc. That's a way better way of explaining the impact of your business.

7:19from a personal level as the founder how did you develop that narrative it's obviously all true but it's a much more powerful narrative than just saying we're ai for procurement and manufacturing how did you develop that narrative and what tips do you have for other founders who are trying to really communicate the size of the opportunity and the impact of their business particularly if they're seeking venture money which needs that building a company is so hard that having some proximity to the problem area and have possibly even felt the impacts of when things go wrong can make a really big difference.

7:55So we were lucky enough at Magentic that this was true for both me and my co-founder, Oren. I'll start with his example first. So Oren saw during his time at OpenAI that it wasn't necessarily access to medical knowledge, the ability to ask questions to these chatbots that was holding healthcare back, especially in rural or hardest reach areas. it was often actually access to medications. And then in the early 2020s, Australia had an issue where actually they ran out of a blood clotting medication, warfarin, because of supply chain shocks. And that affected many patients there on the ground. So Aaron, coming from a medical background, had these two real images burnt into his mind of, we can build all this stuff in the digital world, but often it's the physical world where actually the biggest impact is, and where the most work is still required to get things up and running.

8:45Similarly for me, I was at McKinsey working with a whole range of large manufacturers and other large companies across all different topics. And I remember we'd start off every year with, oh, okay, wow, this really is the biggest procurement shock of a generation. First, it was COVID, then it was war in Europe with the war in Ukraine, then it was tariffs, then it was the Strait of Hormuz. And it really felt like every single year there was this catastrophic supply chain event that would shift everything around. And sometimes it wasn't. Sometimes it meant we couldn't get access to toilet paper.

9:13And sometimes it meant we couldn't get access to our essential medicines. So as a place to go and work, I think finding somewhere where we can work on the most tangible and real world physical impact areas of AI was something that really excited Erwin and I as we're starting to think about the space. You mentioned that the sort of physical world aspect to what you're doing. I think as an investor, I'm super keen on companies that bridge software, humans, the physical world, robots, sensors, you know, because that stuff isn't going away. There's a world where Anthropic is the last company on the planet.

9:49But, you know, humans are still going to exist. Sensors, robots, all of these things in the real world are still going to exist and people need to produce things that we touch and feel. So I think it's a great focus area, but you are dealing with super sensitive industrial data for some of the biggest companies on the planet. And I wonder how you deal with this tension between privacy and data sovereignty for these companies and the ability to take their data, train your models, improve the product for them. There's two things that are important. Firstly, is there's all of the typical stuff that gives big enterprises comfort that even though you're a young startup, we only launched in July last year, you've got everything that they would expect to have.

10:34So everything you've got around your SLAs and you've got the ISMS set up correctly so that you've got the information security and the data handle and you've got the right encryption and your ways of processing and handling data are fully, fully secure. So some of the ways that we make that a bit easier at Magentic is we say everyone has a single tenant deployment on a specific cloud. We often deploy on the clouds of our customers' side themselves, sometimes via a direct link so that no data even touches the open internet. And little things like that can make a really big difference. And then there's the AI side.

11:05And I think very much the CIOs we're working with are still trying to understand the full implications of AI. There's a lot of dark AI being used at the moment where people are dumping big amounts of really sensitive data into open models or the closed models that are provided by some of the US Frontier Labs. And what we often can make a really strong case on is your procurement and supply chain teams desperately need to be using this technology to stay ahead of this wall of demand that they're facing. You can present them the option of, hey, do not use any AI, we're completely going to shut you down.

11:35And then what we see at every single company happening is this dark AI. Or we can say, actually, look, we're going to go through with the data processing agreement. Here's every single AI model we're using. Here's how we use it. Here's the zero data retention policies that we have with them. Here are the smaller little models that we've built ourselves, little things like embeddings models, so you can speed up data retrieval and access in different places. There's no training on any of your data. Your data stays yours. If our AI agents learn from you, then those contextual improvements stay within your single tenant deployment.

12:05And that gives all of our customers some really, really sensitive organizations who might be involved in defense and government type work, the comfort that actually their data is safe and secure. It's really interesting because we're hearing different things from different companies and you can partly split it by the level of sensitivity of the data in the industry. But there is a case to be made for pushing really hard on customers to allow you to train sort of customer agnostic models to improve the product for everyone. I think the reality is if we look at some other application layer categories like for example legal tech, the quality of the models are dealing with text-based information so the documents that lawyers deal with every day is incredible so Anthropic can release a set of tools which will do very basic legal work and then Harvey and Lagora and those companies will then extend that into a very clean suite of tools for an actual lawyer and you can see that the application layer has really benefited from how good that the foundational capabilities are for things like textual review.

13:12So Harvey and Lagor are both growing at completely insane rates. That isn't yet true in procurement supply chains and LLMs inherently aren't as good at dealing with the physical world. So the kinds of data that we deal with is often tabular, often has cross-linking in lots of different kinds of places, is much more numerical in its form. So much more of the work that we have to do is build up capabilities for AI to be able to operate the frontier level capabilities on tabular formatted data that lives in ERP systems like Oracle and SAP. And so what our customers are often asking for us to do is they care about the outcome most of all.

13:51And all of our pricing models are based on outcomes. So we can deliver something for them when it's outcome-based. Where they're pushing us to share more across different customers is actually how do we turn the capabilities of frontier models and make them relevant for ERPs and the physical world and the data they live with, rather than can we train our own large language models, which aren't well suited to their world anyway. And how many majors do you have now? Sure, we've got a whole bunch of different majors. When we started off, we'd actually give each one a specific name. And for example, people would hire Sam, which was a major oriented around value leakage.

14:26And Sam would even have Microsoft Teams account and would start messaging people within the team because of a really strong philosophy around work where you work. So we bring the mages to sit in the systems that you have today. That's less and less true. So right now we've got something we've even called the arch mage, which again is a reference to mage as a magician and it's kind of an arch mage. And the arch mage is probably 15 to 20 of the mages that would be working and deploying with our customers rolled up into one single very powerful arch mage. And what the arch mage can do is act like a super powered digital worker, where they don't just have the capabilities of a source and category manager or a buyer, but they've actually got capabilities across different parts of the supply chain team.

15:08And I think that's been a better way for our customers to engage with the majors. Yeah. And how do you set it up as a customer? What do you need to give it access to? And how long does it take for it to how long until it can start actually acting on behalf of the organization? So there's a few different levels of, we can call it like trust or integration that we're building up here. The goal for everyone is an autonomous AI agent that has context, the data it needs, access to the systems it needs, and the trust of the users that it's going to be supporting in their work. That doesn't happen on day one.

15:44Many of the companies we're working with, they don't have any AI or they've got a very basic Microsoft Copilot subscription. So firstly, has a lot of trust building. The journey that we take our customers on is we say, hey, rather than run a very typical, let's do a big RFP sales cycle into a pilot, into a staged rollout, how about you just start an internship and hire one of our digital workers. That digital worker can join you on Monday next week. It's not going to be able to do everything and do everything all at once, but we'll bring a full deployed engineer to expand the capabilities and the impact of that digital worker over time.

16:19And so what we see with our customer base is that the first few weeks is getting the digital worker onboarded, access to the right kinds of systems. We start with just read access or even just dumps from a data lake so that we start to build up that contextual knowledge. And then by about month two, so four to five weeks in, that agent is ready to start taking real actions. And that can be something as simple as sending an email to a supplier with the context that it's now built up in the human levels of approval through to making small updates to what we call master data records or the core kind of systems within ERP.

16:54And then it can start running things like negotiations, making small adjustments or recommendations when a purchase order comes in that it's going to the correct supplier. Little pieces like that start being built up so that by the time we get to month three, we have a lot of trust. People have seen it working. They've spoken to it. They've seen the actions that it's been taking. We have a lot of work on audit and visibility within our platform. So really month three, we've got with many of our customers, an AI agent for the first time running large parts of their procurement supply chain. Awesome.

17:24And for the layman who's not an engineer, how is an agent actually built? Is it just code or is there a platform that you use to build agents? If someone listening wanted to build an agent, how would they go about it? Firstly, to flag, I'm not technical myself. My backgrounds at McKinsey and Company. So my space is understanding my customers, their language, their problems, the product that they need built. And then I've got my incredible CCO who leads an incredible team of AI engineers and researchers who go and actually build our products on agents. What I'd say at the highest level, and when I'm talking to my customers and trying to bring them on the journey of what an AI agent really is, is LLMs imbued with the ability to take some actions within systems.

18:07And the way that they take those actions is they might have some pre-programmed APIs. So they can send some data to an API or pull it back. They might be able to access some files that you could store in a messy repository because LEMs are very good at dealing with that kind of mess. They can update those documents, they can move pieces around, and they can talk back and forth via an API. That'd be the most basic way that we start to bring our customers on that journey. Of course, some of the most advanced agents we're seeing in production now, and we love to use as well, are starting to use computer use agents.

18:38So to what extent can we actually have an AI that looks at the screen we have on our laptop, whether that's a website or a piece of software like SAP, understands the screen, understands the goal that it wants to achieve, like to find a new supplier, and can actually click like a human being would. And it does that by sending screenshots, finding the right place to click, and then honing down on it. So that's kind of the full spectrum of AI agents. Very cool. Thank you for explaining that. And then it would be amiss of us not to shout out sequoia and um our friends there and you're backed by sequoia capital so can you share a little bit about your experience of raising money from a fund like them and tips tricks of fundraising playbook and anything that you've got to a listener who's thinking about fundraising firstly i think sequoia capital incredible partners and julian beck who sits on our board is an incredible thought partner as well we were on stage together at raise ai summit in paris a month or two ago and we were just having this great conversation and even just then just sparring backwards and forwards on stage in front of people.

19:41It's amazing to have somebody really thinks about our company almost at the same level of depth as I do. And I really see that with all of the Sequoia team. In terms of what that fundraising process was like, we actually took part in a program called Sequoia Arc, which is a kind of a mini accelerator program. It's five weeks long, but you get a full size kind of seed type funding package with it. So you get the benefit of the Sequoia with the capital and then a five week accelerator as part of that as well. Oren had met one of the talent partners at Sequoia at a previous event while he was still in research.

20:12And then through that, when Oren and I were working over the summer, try to figure out something interesting to go and do. We had a few conversations with the Sequoia team. I think they had a thesis that there was going to be something valuable in the physical world. And through that, we had a whole bunch of meetings, probably had like six or seven different partner meetings with Sequoia team. At the end, they decided to invest. We're lucky enough that they've invested in every round we've done since. And maybe one more tip I'd add as well is I think it's a great idea for founders to think about the entire package of investors that they're bringing together.

20:41So Sequoia are incredible on the company building, the technology side, partnerships, all those areas. They're incredible. We really wanted to supplement them and complement them with deeply technical on the manufacturing side of investors as well. So as part of that, we brought on board the Wesley Group and First Mountain Ventures, who are two funds that are all of their LPs are the big manufacturing companies in the US and Germany. So what we get is that way we get access to customers, we get access to those learnings, we get this DP technical partners joining as well. When we were thinking through our most recent funding round, which we announced a few days ago, that was led by Felicis and the incredible Sundeep, who's one of the founders of Felicis, joined our board.

21:21Again, that was a technology and company building bet. So it's always a balance of bringing both partners in so you can build the best company possible. Yeah, incredible. It was great to see the fundraiser announcement. So well done to you guys. And from your vantage point, what kind of impact do you see your company and other companies in the physical world having on the world around us? Because it's hard to picture exactly what it means for the rest of us. Like, sure, there are companies that are improving margin a few points here or there. There are companies that are allowing people to do eight hours work in seven hours.

21:55But when all is said and done, when your version of the world exists, what's so different that you're excited about? I think there's two things. There's the reality. And we're all based out here in Europe. And stuff is too expensive. So food through to medicines through to physical goods. There's a cost of living crisis. And it's been compounded by these once in a lifetime shocks that we've already spoken about. The fact that it's so expensive is partly the cost of manufacturing things and the cost of the labor, the government and the regulation, the environmental requirements for these companies as well.

22:29And all of this is compounded in your supply chain. So to the extent by which we can help companies shave off percentage points of cost, increase by percentage points the speed, find more win-wins with their suppliers where actually, hey, the waste of one supplier's process could be used by another one of your suppliers. And those two would never speak. But now because you've got a better organized procurement supply chain, you can actually find those win-wins. all of that what it really drives to is a more affordable world that we live in physically which can make a really big difference in for us today we work with three of the top 10 food and beverage companies in the world doing exactly this and then we work with many of the world's largest technology manufacturing companies as well and for them it's actually hey if we know that we're going to have 30 percent more data centers in two years time than we have today and we need that to build the drug discovery research pieces that we'll work on over there.

23:26And we want to launch new rockets. And we want to figure out how we can get a supersonic flight to happen again. For that as well, the faster we can get our supply chains to move, the more we can work into places like new product innovation. So I think we're lucky at Magentine that we get to do both what's really grassroots through to some of the more advanced stuff as well. Yeah, Robin, and in categories where there is competition, that will filter through to the customer because they'll have to price down as their competitors do as well. In categories where there isn't much competition, could that just mean better margins for the companies?

24:00Will they just be more efficient businesses and not really pass that through to customers? I'm an economist by training. I think that the higher that the profits become in an industry, the more of an incentive it is to go and enter that industry. I think for a long time, it's been difficult to enter new industries. If you can imagine how hard it is to start a new car company in the US today. And I think partly that's because if I want to start a new manufacturing facility, I need to go and I need to build relationships with 5 ,000 suppliers, build up the credit with them, build up the supply chain to be able to go and manufacture something.

24:29To the extent where we can deliver higher profits to those companies, make it a more attractive way to enter, and then also make procurement supply chain a more manageable area to work with, that should hopefully spur more competition, which is great for new product innovation, it's great for customers, it's great for the companies that get to compete on that plane as well. Yeah, and I guess with AI, the barriers to entry are disappearing for lots of different industries. If someone gets too greedy, their margins are too healthy, someone else is going to come in and help bring it down for consumers.

24:58That's the good side. Awesome. Okay, so Robin, it's so interesting. It's really great to hear more about Magentic and your journey as a founder. We're going to move on to our final two questions. The first is a future unicorn prediction. So have you seen another company that you think has the potential to go all the way through to unicorn? Yeah, sure. And I was thinking a bit about this earlier today, but there's a company called Alike Labs. And what they're doing is they're doing work coordination, but from a social coordination perspective. And if you can imagine what it looks like, they sell to companies like mine.