#743: The Frugal Architect w/ Werner Vogels: The Ocean Cleanup's mantra: Start simple and iterate relentlessly

0:00This is episode 743 of the AWS podcast, released on October 27th, 2025. Hello, everyone, and welcome back to the AWS podcast. Simon Lees here with you. Great to have you back, joined by two very special guests. And this is a special episode, one of our Frugal Architect episodes, which I know a lot of you really hang out for. And of course, you can't have a Frugal Architect episode without our VP and CTO of Amazon.com, Werner Vogels. G'day, Werner. How are you doing? Hi, Simon. Yeah, I'm doing well. I'll be having a good time here. It's good to have you back to talk about the very important topic of Frugality, and to help us with that is an amazing customer guest.

0:40We're going to be speaking with Boyan Slatt. Boyan founded the Ocean Cleanup at age 18 with just€200 to€300 of saved pocket money after encountering more plastic bags and fish while diving in Greece at the age of 16. That's confronting. What started as a viral TEDx talk in 2013 has grown into a global operation that's removed over 70.5 million pounds of trash from aquatic ecosystems worldwide. His approach to engineering is defined by constraints-driven thinking, so he's in the right place, starting with limitations rather than capabilities, iterating relentlessly and letting mission guide technical decisions.

1:19Under his leadership, the Ocean Cleanup has developed systems from passive ocean collectors to AI-powered river interceptors operating across 20 rivers in nine countries while targeting 1 ,000 rivers by 2040. Boyan, welcome to the podcast. Thank you for having me. That's a hell of a journey. Yeah, and actually, I wanted to say, we've actually, I think we're at about 91 million pounds of trash collected now. I was going to say, but who's counting? But clearly, it's important to count. That's right. Well, you know, you were quite sure when you started this, 18, a Delft, I assume, you were still there.

1:59So tell us a bit about the story. What drove you to get this going at all? Sure, yeah. I think I've always been a bit of an inventor all my life. I've always been very passionate about building things, engineering. Going all the way back to when I was two, three years old, I remember building my own chair out of wood and nails just because I thought that would be more fun to sit on than one my parents would give me. And then it went into my next passion was computers. So I assembled my own tower and then I was into chemistry and I was building explosives and smoke bombs. And one day I almost burned down the house because I was distilling ammonium nitrate on the stove.

2:59I kind of lost track of time. Yes, you do. Yeah. And I kept the entire house in smoke, which was my mom really loved that phase. And then I was into model rockets. And then when I was 12, 13, I decided to set a Guinness World Record, launching more than 200 of them at the same time so uh so i've never been bored is the the short summary there um but it all wasn't very useful it's just sort of having an idea and then making that real i think there's just an immense amount of satisfaction i get out of that um but then yeah when i was 16 i went scuba diving in greece and i was hoping to see beautiful nature i went underwater and I came across more plastic bags than fish.

3:54And so the natural thing for my brain to ask myself was, well, why can't we just clean this up? And once an idea like that or a question like that start brewing in my head, I really cannot stop thinking about it. And then I went to to Delft to uh to study aerospace engineering uh but yeah at the same time I still was obsessed with this this idea and then after half a year um there's this point when I said like well I can't really do both so let's give this a try um if it doesn't work out it can always re-enroll after half a year uh but then you know a few months later there was a TEDx presentation I had given that went viral, was shared millions of times, and then allowed me to build an initial team, raise the first few million dollars, and, yeah, so it got us going.

4:53So, yeah, that's... So, talking about funding there, originally crowdfunding, but so how did you get from crowdfunding crowdfunding to having a, let's say, sustainable income for your business such that you can continue to run it? Of course, the challenge, what we're trying to solve is essentially, you know, we're creating a tremendous amount of value. So when you look at the harm done to ecosystem services by plastic, it's estimated to be between 500 billion and two and a half trillion a year. So, and there are not many companies with that level, if any, that level of revenue.

5:40But of course, most of that value is not value that's captured in the GDP of human civilization, right? It's things like clean air, coastal protection, you know, being able to go to a coast and clean, having clean beaches, having marine life, all of that. So, of course, we tap part of that through tourism, fisheries, etc. But a lot of it is not captured. So I think if something can be a for-profit, it should be a for-profit because then there's such a beauty in having this alignment of incentives between your team members, shareholders, wider society. but the problem is of course when you're doing something like this which is really outside of the market you don't have that so that's why I opted to start a non-profit when I started this I really didn't see any other way to to get it off the ground and you know crowdfunding actually the first crowdfunding campaign was$89 ,000 and then the second one a year later was$2 million dollars um but um you know that that got us started but of course wasn't uh of the scale that was required to really make this work and but thankfully over time i think the way we've approached this is um is almost like investment round so basically we raise some money and then we get results and then with those results we get to the next level of credibility and And I think your level of credibility has to be proportional to the amount of money you can raise.

7:21Or it naturally becomes that. But you're also building things. So for that, you absolutely need capital to be able to do that. Yeah, exactly. So I think then over time, we were fortunate enough that both individuals and companies stepped up to support us. So you got people like, you know, Mark Benioff, Salesforce, Joe Gabbia from Airbnb, as well as literally millions of other people who decided, okay, I want to be part of this. This is like an Apollo project for the earth. This is really exciting. And we want this to happen. And then over time, of course, at the beginning, it was this very high risk, high reward project.

8:05but over time as risk came down and now it's at this moment it's not a question more whether we can do it we know exactly what to do it's really just a scaling challenge it yeah companies also started to get involved and there of course it's more than philanthropy it's really an exchange of value right because for corporates partnering with us it you know it allows them to have this variable story, increases brand value, increases awareness. So these partnerships is something I'm really excited about. And the journey is a business type of transaction. Recently you heard a lot about microplastics that are inside people and inside fish and things like that.

8:52Is that a relationship between what you are trying to clean up and those microplastics? For sure, yeah. The fortunate thing is that only a very small fraction of plastic in the ocean is currently microplastics. It's in order of a few percent. Most of the plastic is actually large stuff. And of course, what we do is, you know, unfortunately, the tiniest pieces, they're really too small to extract. Unfortunately, over time, you know, they will wash up on shorelines, so they won't be there forever. But of course, by removing everything that's not microplastic, all the larger objects, you prevent the creation of those microplastics because they are really coming from the degradation of larger plastics.

9:38And then on top of that, by what we do in rivers, we prevent the inflow of new plastic into the ocean. So yeah, so I think what we do is we really prevent this ticking time bomb of the amount of microplastics increasing tenfold, hundredfold in the next few decades if we don't clean this up. So, Brian, let me take you back to the early, early, early days and your first system called Wilson, which I understand didn't last too long before it sort of reached its limit. Tell us about that and what that really, how that changed your thinking. So the original vision to clean up the Great Pacific Garbage Patch was to have a giant U-shaped barrier that was passive, that was just free-floating.

10:24It would be pushed forth by wind and waves and would corral the plastic inside of it. So we'd only have to come by periodically and empty it out and bring the plastic to land for recycling, which was a very elegant idea on paper. But in reality, it turned out it didn't collect plastic. So we took it out, then we deployed it, and then they didn't collect plastic. And really just a few months later, it broke into two. And obviously, we learned a tremendous amount from that. And yeah, after that, we completely changed the approach. And then we eventually got to assist in that work. But yeah, looking back, I think what I would have done differently is that obviously we learned a lot, but I think we could have learned the same lessons in a cheaper way.

11:26We truly believed that that was the system that was going to clean our ocean. It was almost like this tunnel vision. And then ultimately, I mean, you're going to fill along the way, right? When you're doing something new, you're going to have so many failures. But I think, you know, your job as an innovator is really to make sure that you can recover from those failures and that you make mistakes in an as cheap and fast way as possible. So it was great what we learned, but I think we could have learned it with a system that was maybe a tenth of the scale and maybe close to shore rather than actually in the mail of the Great Pacific Garbage Patch.

12:04So that's kind of the thing I would have done differently. So you started off with trying to attack the biggest problem there was. Yeah, right. So it wasn't really this MVP approach, I would say. And before that, we had done scale metal tests and we had done prototypes near shore. But the actual thing of catching the plastic, which is, again, quite an important requirement for an ocean cleanup system um we um yeah we we only really put that to the tests at full scale in in the real environment while you know maybe we could have done that with simulated plastic or something closer to sure so these are these are lessons where there's a famous t-shirt i'm sure you've all seen that says uh i don't often test but when i do it's in production and that's sort the experience yeah and i think you know when we when we started out we really didn't i think we went through six or seven conceptual iterations over the years and it was really like um the way imagine it is like a land like a landscape full of hills and your goal is to climb the get on top of the highest, the tallest hill, and then there's this dense fog, and you can only look three to five meters ahead of you.

13:28And at some point, you know you're climbing a hill, but you don't know yet whether it's actually going to be the right hill, the one that you actually want to end luck on. So it does take a certain degree of trial and error to find that hill. But again, it's really just about trying to limit the cost of that experimentation. Absolutely. It's always very hard to know, am I climbing the right hilltop here? I think one thing that I noticed is when you're actually climbing the right hill, things seem to get simpler and simpler over time. while when you're climbing the wrong hill, actually things get spiral onto complexity.

14:19So I remember when we were still trying to make that passive system work, there was this challenge of these contradicting design requirements where on one hand you want the system to be flexible, to be able to follow the waves so that it doesn't break under fatigue loads. But on the other hand, you also want to keep the system open because you don't want this view to collapse because otherwise plastic can't enter, right? So then at some point in time, we got to the point that we were doing engineering studies on literally an Eiffel Tower-sized steel structure that was submerged below the water level that would act like a spreader to keep that U open.

15:02And then at some point, I was thinking, guys, what are we doing here? this is insane we're creating a literal eiffel tower here uh like a steel truss uh structure and and um yeah and then it's it's kind of this um you might be familiar with this idea of epicycles you know before copernicus figured out that everything was rotating around the sun there was there were these other models um that kind of mathematically you could make work but basically You got these little circles and circles, and it just was not simple at all. But you could kind of make it work. But I think that when you see EpiCycle, it's kind of a hint that, okay, I'm not on the right track here.

15:48I'm climbing the wrong hill. So in a lot of the early systems in the sort of starting off point, you were using a lot of sort of off-the-shelf hardware and open-source models for computer vision, et cetera. And things have changed a lot since then. but what did those experiences teach you that you still apply today? Yeah, so computer vision, general sensing, is really a core part of our solutions. If you think about it, the key to our success is being there where the plastic is. that applies to both what we do in the great pacific garbage patch in the middle of the ocean as well as in rivers where you know in in this great pacific garbage patch you got an area three times the size of france where on average the plastic is super dispersed but then that you got when you take a sailboat through it there are days when you see hardly any plastic and there's days when you see tons and tons of of plastic around you so being able to determine where these hotspots are, which by the way are dynamic.

16:58They come and go all the time in different places, different times. That is really the key because otherwise you're just basically sieving the ocean. It's really about pinpointing those places where the plastic is. And similarly in rivers, what we found is that less than 1 % of the world's rivers is responsible for 80 % of plastic pollution. So being in the right, in the correct rivers is the key to being able to do this effectively. So what that means is really when we go into these coastal cities, it's deploying hundreds of sensors around the city, being able to determine which are the, you know, what's the pretter here, which rivers do most of those emissions and where in those rivers should we deploy.

18:01And, you know, for example, some rivers, you've got vessel traffic as well, right? So you can't block off the entire river with our systems. So there we released GPS trackers that mimic the flow of plastic. So we can really, again, strategically deploy there where the plastic goes while giving enough space for boats to pass. So it's all about being at the right place at the right time. And even though the cleanup system, the hardware is simple and it must be simple because complex tech doesn't, complex hardware is very hard to scale. It breaks down. It's hard to maintain. It's hard to deploy, et cetera.

18:41So you want to keep the hardware as simple as possible. And then basically all the complexity, you know, that's really applied to more of what we call the software side, right? So really the intelligence of, especially up front, in terms of determining where you deploy. So why is it actually that these 20 rivers are the ones that are causing most pollution? I mean, is that because, is it humans? Is it factories? is it? I mean, how come there's only these 20 rivers that actually take 80 % of the plastic? Yeah, well, so we are currently deployed in 20 rivers, but the 1 % actually represents in the order of a thousand rivers.

19:23So it's still a significant amount. But I guess when we say rivers, a more accurate word would be waterways. Most of them are pretty small drains. And And what we see is that these rivers are concentrated in coastal cities and middle-income countries where you got a combination of a high concentration of population close to the coast. And you have basically very poor waste management infrastructure. These are countries that are rapidly developing economically. people now have enough wealth to buy a lot of stuff wrapped in plastic, but the government basically does not have the money yet to do proper waste collection and disposal.

20:08What's quite interesting is that there's really no correlation between the amount of plastic that a country consumes and the amount of plastic that it emits to the ocean. Essentially there's this thing called the CUSNATS curve, which is an inverted U, where the poorest country, they hardly emit any plastic because they don't consume any plastic. The richest countries, they consume most of the plastic. So the world's richest countries, they consume about a third of all the plastic, but they're responsible for less than half a percent of global plastic emissions because here in Western Europe and US and Japan, Korea, we of course have good systems to collect the waste and dispose of.

20:55It's really this middle ground, places like Jakarta, Manila, Mumbai, Accra, a place like that where you have a lot of people, a lot of plastic now getting consumed, but basically no way to collect and dispose of it. Yeah. I was very fortunate a few weeks ago in the Amazon. Oh, yeah. No plastic in the Amazon. I think what's quite interesting as well is that most plastic that ends up in rivers never makes it to the ocean. So whatever plastic gets, we actually tested this again with a GPS tracker experiment. So a year ago we released I think about 100 GPS trackers in Manaus in the Amazon. None of them made it to the ocean.

21:48and that's really because it's so far inland and of course the slope of the river is also it's very shallow, right? It's like an ocean almost. It's not like a very aggressive, fast-flowing river. You know, that plastic just gets sort of beached on the riverbanks and gets stuck in twigs. But that stuff, yeah, really doesn't make it to the ocean. So, yeah, so that's, of course, also, you know, when we started out back in 2017, our first global river model actually predicted that the Amazon would be a key contributor. But back then, we didn't really understand the physics yet. We didn't know this fact.

22:34And now when we take that into account, actually, the Amazon just completely drops off our list of priority. which is good because it's an insanely big verb and I'm very glad we are to do that one. So, Brian, tell us a bit more about something the team did using some artificial intelligence where you cut data transmission costs by 99.975%. That's a lot of cottage. Can you walk us through how you used AI as a practical tool rather than a silver bullet? We developed something we call ADIS, the Automatic Debris Identification System, which are these cameras that we attach to ships, just merchant vessels like cargo ships, container ships, that go around the world all the time.

23:26And whenever they enter port, they send a data package to headquarters here. And it actually tells us the concentration of plastic along the path in the ocean that they traveled. And we use this to better determine the location of the plastic in the Great Pacific Garbage Patch, but also as a global thermometer for how we're doing. because ultimately we make it very explicit. We want to help ourselves out of business. But for that, we need to know when we're done. And we really need to be able to track what is the trend in terms of how much plastic is in the ocean. So by having these ships continuously traversing the ocean, getting that data, we are able to have this global plastic thermometer.

24:20and uh so these models run on the cameras itself so all the processing is done locally at the edge and you basically get the data the resulting data from that back amazing so that's such an impressive way of doing things now one thing you touched on here is is the global nature of the work you're doing um which introduces a whole lot of extra complexity i mean we talked about um 20 rivers, 9 countries, you've got ships going everywhere, there's a lot of stuff going on. How do you design your systems so that you don't need a specialist engineer everywhere? Because I can imagine that could get quite expensive if you went down that trail.

25:00Yeah, so I guess that's another one of those items for the list of mistakes I made. which is the original system that we developed to intercept plastic in rivers was a really cool machine. It was fully solar powered, fully autonomous and like a trash eating robot essentially. And then we deployed them in places like Thailand, Indonesia. and then you discover, oh, well, actually, first of all, it's very hard to import those things because how do you classify it? Is it a boat? Is it a barge? What is it? There's no sort of line-eye. There's no tick box for trash-eating robot. Yeah, and then you need the permits to deploy.

26:00Again, same problem. then installing it requires specialized installation vessels and it's quite difficult then you need to train people to operate it again pretty difficult you need people to you need to have the maintenance and then you have these PLCs on board and then a tiny thing breaks and you have to air freight something from Germany to to fix it and it's just a big mess right so um so then we got to the conclusion that okay maybe people are underrated and uh actually creating jobs in these places uh is uh is actually also something that is a positive side effect to to what we do so so then we kind of backtracked on that and went to to more you know simpler hardware uh simpler ways you know using excavators for extraction rather than you know conveyor belts and um yeah i mean that's that's working quite well of course now the key is so the technology is quite simple but now the we still you know looking ahead right now we we're doing about one deployment per quarter uh next year we want to get to a point where we're deploying one interceptor per week roughly um so that scaling ramp it really requires us to rethink the way we're doing things it's still too much reinventing the wheel every time seeing everything as a unique project so it's all about building this project factory now where we standardize everything from the contracts we set up with the operators to all the individual hardware components to the installation manuals it all has to be standardized because otherwise we will not be able to get to that deployment gate.

28:03You're generating or collecting an enormous amount of data with all these ships and also the things that you have yourself. What do you do with the data? I mean, beyond just running your company and beyond making decisions yourself, is this something that you share with others as well? Yeah, for sure. So actually, we're just going through a review of some of the next cities we're going to tackle with our interceptors this morning. And it's amazing. For basically an entire city, we mapped the entire river network. We know exactly where all the landfills are. We know exactly where we need to deploy the trash flux out of each of the waterways.

28:47It's amazing. And then, of course, even post-deployment, we are collecting tons of data because we're able to see, okay, what's the trash we're collecting, the composition, the quantity. So it's, and of course, we're using this, all this data to deploy the best possible solutions and to further optimize their efficacy and improve the next set of deployments. We're also using it to, of course, convince funders, partners, to show what we're doing and we have a good plan to tackle a given city. But we're also sharing it with the local government because they are actually able to make better decisions with that data.

29:38For example, in Jamaica, one of the things we found is that the bulk of the trash was styrofoam packaging. We shared that data with the government and then they themselves decided to ban those styrofoam clamshell takeout packages in Jamaica just because they saw like, okay, actually the majority of trash coming down the rivers. That's what it is, yeah. So yeah, it's, and I guess a broader point there is that what we're doing, of course it's not the the ultimate solution ultimately we don't want plastic to even flow down these rivers we hope one day those interceptors are not necessary anymore but that's not going to happen tomorrow but it's it's not either or right what we're doing through the data we collect through the visibility we generate for the problem we're actually able to help catalyze that that upstream change as well so do you think you'll ever be dull yeah for sure and I think there's two levels of done I think there is the point when the oceans are clean which I think we should be able to express in years not decades so right now we're stopping between 2 % and 5 % of global plastic pollution we've doubled last year, we're doubling again this year So just mathematically, you don't need that many more doublings to get to 100%.

31:15So, yeah, and I think now with this 30 cities program, we want to tackle a third of emissions in the next few years. So that should allow us to maintain this annual 100 % year-over-year growth. so I think pretty quickly we'll be able to get to the point that hey great OSHAs are clean again at that point however those interceptors are still necessary because if you stop operating them the OSHAs will get polluted again and I think that will be true for a while before the entire developing world is as pristine as Singapore or something, I think will take a few decades. And of course, we need to make sure that the interceptors will keep operating during that time.

32:12Of course, what we do envision is that after a few years, the government can really step up in essentially integrating our interceptors into their waste management infrastructure. So it's basically just like they have garbage trucks in the streets. They need to operate interceptors in their rivers. Interceptors, yeah. So yeah, I think clean oceans very soon. But then the day that we can take out all the interceptors and it's truly done, I think that will be decades away. Now you spoke about the growth rate and how much we need to double, etc. And obviously you want to maintain cost control to make sure you can grow in terms of the collection.

32:59And you've often said that cost per kilogram of plastic removed is one of the KPIs you really most value. Tell us about how you apply that metric to your technical decision making. Yeah, of course, the unit economics are very important. so it's so we use this to really prioritize where we where we deploy and it's also a metric that we use to drive continuous improvement both in terms of initial investment, operational cost it's something we use to judge the performance of our operator so the local contracted company that actually runs the interceptors. So yeah, so it's definitely a key metric.

33:48It's not the only metric. And I think one nuance that we're now recently started making is that it's really not cost per kilo of bus, but it really costs per unit of impact. Because in some places you can imagine, say, two hypothetical cities, one city that emits a thousand tons a year, but where it's basically just sort of mud flats around the city. And then there's another city that maybe emits just 100 tons a year but those 100 tons go straight into a coral reef with a tourism hotspot where, of course, the social environmental damage would be much greater than in the first scenario. So we're now able to actually put a dollar value to the ecosystems around those cities where the plastic goes.

34:43So we're able to better, again, make better decisions in terms of where we should prioritize our deployments. That's amazing. Where does the plastic go? Like you're collecting this plastic. What happens to it all? The places where we go to tend to be the places where, of course, where the biggest leakage occurs and where the greatest leakage occurs tends to be the places where the poorest waste management infrastructure exists, right? So it is always quite a puzzle to figure out, okay, what do we do with this waste? And in most cases, we are able to find suitable, responsible destinations for this waste.

35:26So this can be, of course, there's this hierarchy of waste destinations. So ideally, what's at the top is recycling. So in some places, there's quite a large PET fraction, which is the type of plastic that's most valuable, easiest to recycle. So there, you know, you have a big chunk that gets recycled. Then, and then below that, you essentially got landfill and incineration, where, of course, we audit those destinations to make sure that those landfills don't leak, that the employment conditions are responsible and to really make sure that that waste never ends up back in the environment and is staying there in a responsible way.