The Exchanges

Q And I said, what should I ask Mathieu? And he said, ask him about the McDonald's test. And I was like, Right. Can you, can you help me understand that? And he said, no, you're gonna have to ask him. And so I'm like, okay, what, what's the McDonald's test?

A In the early days, and we still do it, like, in the street, uh, we would go with Elliot, and we would go to McDonald's, and literally we would go in the line and offer people, like, willing, ordering things, ordering food, like, can we pay for it? And you just test photo room. And we would run, like, Yeah, 10 questions about photography, and then we would, like, let them try the onboarding and really give us feedback, and every time they wouldn't get to the new feature we're trying to do, like, it was a new feedback, we discovered something, and it was the best learning experience. It actually comes from the, uh, Zenly team, they were doing that in Paris. I did it so much that I, I got banned from the McDonald's from, uh, next to our office, so it's just, like, I couldn't enter it anymore, like, the security guard would come to me and say, you can't, you can't get hidden, like, you can't. You can't go in because you, you're bothering your customers.

AI assessment note: “we would go to McDonald's, and literally we would go in the line and offer people”

Q And I said, what should I ask Mathieu? And he said, ask him about the McDonald's test. And I was like, Right. Can you, can you help me understand that? And he said, no, you're gonna have to ask him. And so I'm like, okay, what, what's the McDonald's test?

A In the early days, and we still do it, like, in the street, uh, we would go with Elliot, and we would go to McDonald's, and literally we would go in the line and offer people, like, willing, ordering things, ordering food, like, can we pay for it? And you just test photo room. And we would run, like, Yeah, 10 questions about photography, and then we would, like, let them try the onboarding and really give us feedback, and every time they wouldn't get to the new feature we're trying to do, like, it was a new feedback, we discovered something, and it was the best learning experience. It actually comes from the, uh, Zenly team, they were doing that in Paris. I did it so much that I, I got banned from the McDonald's from, uh, next to our office, so it's just, like, I couldn't enter it anymore, like, the security guard would come to me and say, you can't, you can't get hidden, like, you can't. You can't go in because you, you're bothering your customers.

AI assessment note: “we would go in the line and offer people... can we pay for it? And you just test photo room”

Q Now, I would love to start. We see PhotoRoom today. What was the aha moment for you that, like, I'm going to dedicate many years of my life to creating this product?

A Well, first, I've been working on photo video apps for the past 15 years. Like, I started that on campus at Stanford during my Master. So it's been a long, long journey. That's why I got in iPhone development was photography in the first place. The aha moment for PhotoRoom is I Was at GoPro that has acquired, uh, the startup I was working at as a head of product. And while we were doing an iPhone app quick for video editing, we're being featured by Apple. And I don't know if you know these things, but they ask you like, can you provide the artwork for tomorrow or yesterday or like very fast? And while we were at this small startup being acquired, but the U S where they were sleeping, our designer was off and tell myself, well, I'm good with Photoshop. I'm good with computers. Let's do it. And I opened my computer, and, well, I ended up spending the afternoon literally, like, trying to get what I wanted. I had a clear vision in mind, and it just didn't work. Like, too many options. I didn't manage to do something I wanted. It was frustrating, and I, and at the same time, I had, like, on the other part of the office, had the AI team, and I knew they were working algorithms that were so helpful, and it kind of clicked for me. Like, okay, we need something easier. It needs to get easier. Programming got easier, but editing didn't, and let's build something. There's an opportunity h…

AI assessment note: “it kind of clicked for me. Like, okay, we need something easier.”

Q What were the other two pillars? You mentioned the first about the shipping, iterating.

A Yeah, the first one is, um, challenge toward progress, kind of, you know, candid feedback, like, if there is, like, you should be able to challenge anything in the company if you think it's not going great, or even, like, even if it's not your field. Second one is instant learning, so really this idea of shipping and learning from that, and how, like, really we're all learners. And the last part is be, be an artisan. We are all makers, and we really love, like, people that master their craft and are making, or crafting, and are the best at their, their craft, sharing, sharing that part with others, learning from others, so yeah, be an artisan. And that's the name of the entity in France, so.

AI assessment note: “Second one is instant learning... And the last part is be, be an artisan.”

Q What are the big fuck ups that you made with the first two that you've done differently this time?

A So I started two companies and joined a third startup that was acquired. So it's kind of my fourth startup story. I think first one is focus. So really, my second startup, like we were doing too many things. We tried to do two products. We were not focused. And one thing that worked amazingly well for, for Photorum is With Elliot, we just, like, from the beginning, every six months, we would focus more. So we started with video and photo, and they say, like, okay, let's just do photo. And then we started to say, okay, let's focus on people selling online. And every time we focus with the next number of people, like, signing in, being happy, being, like, very engaged with the product. So I think this one was very, yeah, very helpful.

AI assessment note: “my second startup, like we were doing too many things. We tried to do two products.”

Q What are the big fuck ups that you made with the first two that you've done differently this time?

A So I started two companies and joined a third startup that was acquired. So it's kind of my fourth startup story. I think first one is focus. So really, my second startup, like we were doing too many things. We tried to do two products. We were not focused. And one thing that worked amazingly well for, for Photorum is With Elliot, we just, like, from the beginning, every six months, we would focus more. So we started with video and photo, and they say, like, okay, let's just do photo. And then we started to say, okay, let's focus on people selling online. And every time we focus with the next number of people, like, signing in, being happy, being, like, very engaged with the product. So I think this one was very, yeah, very helpful.

AI assessment note: “first one is focus. So really, my second startup, like we were doing too many things.”

Q Can I ask, do you think the speed of execution is as important as everyone says?

A Yeah, definitely. Like, actually, it's, so we have three cultural values at Photoroom. One of them is, like, uh, instant learning, and literally, like, speed of execution, like, ship, learn, iterate is one of them. Like, You know, you have a, like, finite time in life, anything like life, like a career, a path in a, in a company, and the number of things you can learn really depends on how many loops you can get in that. And I really want, like, everyone working with us to, like, think about it. Like, how can you learn faster from what you do, from what you ship? So, for us, it's critical. I think Elliot, my co-founder is, like, he's an amazing developer, but he's, like, the fastest builder I ever met. Like, he ships so fast, it's amazing.

AI assessment note: “Yeah, definitely. Like, actually, it's, so we have three cultural values”

Q What were the other two pillars? You mentioned the first about the shipping, iterating.

A Yeah, the first one is, um, challenge toward progress, kind of, you know, candid feedback, like, if there is, like, you should be able to challenge anything in the company if you think it's not going great, or even, like, even if it's not your field. Second one is instant learning, so really this idea of shipping and learning from that, and how, like, really we're all learners. And the last part is be, be an artisan. We are all makers, and we really love, like, people that master their craft and are making, or crafting, and are the best at their, their craft, sharing, sharing that part with others, learning from others, so yeah, be an artisan. And that's the name of the entity in France, so.

AI assessment note: “challenge toward progress... Second one is instant learning... And the last part is be an artisan.”

Q I totally get you. Kamal, what worries you most when you look at the future of the space? What, what concerns you?

A I think, as you say, like, the field is moving, uh, very fast, and AI is making A lot of product that kind of obsolete. So you have to think of like, to make sure you're building like a old product with new tech. I think that's a very important part. To give you an example, the first company I made, I was trying to make like postcards and print them from my smartphone. And while looking back, the obvious answer is like, okay, that's not what you want to do. You want with the new smartphone, you want to create Instagram. Like you want to share an image of where you are and share it with your friends. That's, like, the new product with the new tech, and I think it's very important for Photorum, like, when the tech is moving that fast to, okay, what's the new way of doing that, not, like, the old way with this tech, and so you need to, yeah, destroy your product all the time.

AI assessment note: “to make sure you're building like a old product with new tech.”

Q Do you think founders today are using new tech, being AI, to solve old problems?

A Well, I don't think, like, doing that exactly works, I mean, especially for new founders, work at all. I think what's happening with AI is you have, like, All past products that are building tech to serve the, their existing customers. So if you take an example, like this is the idea of the, the co-pilot, right? I mean, if you're an incubant, you're, you're a car maker, you're going to build like the navigation part, you're going to build like a, some security features, but it's just on top of what's already existing. If you're the, the, the newcomer, if you're the founder, you're building the Tesla, you're building the, the autopilot part where The input is like, I want to go to this place, I want it to be like a scenic drive, and boom, yeah, it gets it done.

AI assessment note: “I don't think, like, doing that exactly works, I mean, especially for new founders”

Q I'm sorry, I'm naive here. What questions are you asking when you sit down with them?

A Two things. First, you want to understand, like, how they approach your, your vertical. So for us, it's great. Everyone does photo, and it's a big asset. And so I would ask, like, what, what, how do you capture your photo? Which app do you use? Like, what do you use them for? Are you using them for fun, for selling, for other stuff? And give you, like, kind of an insight. People get comfortable. Well, it's really show don't tell, like you show them the app, they download the app, and well, it's very difficult because you get frustrated because you know where they want, you want them to tap, but all the value of it is like them telling you how they feel, what the, what do they understand from the app, what, uh, they wish was happening now, and really understanding what's, uh, it's user testing, like really understanding what's the process for them when they see that as a first experience.

AI assessment note: “I would ask, like, what, what, how do you capture your photo?”

Q How do you feel about the co-pilot strategy? I had someone on the show say it's the last dying breath of incumbents, which I quite liked.

A That's quite, yeah, it's, I'm quite aligned with that. It's quite brutal for them, and I'm sure some will survive, but I think the problem is where I see that being very true is if you have an audience today, like you have programmers, you have designers, well, you want to, you don't want to alienate them. You want to give them extra value, so you're kind of thinking, what can I build for them that makes their life easier, but you don't want to, like, Have something that's kind of an autopilot to replace them. That's why I take the, like, yeah, the Tesla analogy. Like, you're making planes, like, you bring a co-pilot technology, but who's buying the planes, the airline and the pilots, and you're not going to, like, sell an airplane with no, no pilot at once. They're going to, like, uh, like, go on strike and everything. So I think the autopilot...

AI assessment note: “I'm quite aligned with that. It's quite brutal for them”

Q I'm sorry, I'm naive here. What questions are you asking when you sit down with them?

A Two things. First, you want to understand, like, how they approach your, your vertical. So for us, it's great. Everyone does photo, and it's a big asset. And so I would ask, like, what, what, how do you capture your photo? Which app do you use? Like, what do you use them for? Are you using them for fun, for selling, for other stuff? And give you, like, kind of an insight. People get comfortable. Well, it's really show don't tell, like you show them the app, they download the app, and well, it's very difficult because you get frustrated because you know where they want, you want them to tap, but all the value of it is like them telling you how they feel, what the, what do they understand from the app, what, uh, they wish was happening now, and really understanding what's, uh, it's user testing, like really understanding what's the process for them when they see that as a first experience.

AI assessment note: “I would ask, like, what, what, how do you capture your photo?”

Q Can I ask, do you think the speed of execution is as important as everyone says?

A Yeah, definitely. Like, actually, it's, so we have three cultural values at Photoroom. One of them is, like, uh, instant learning, and literally, like, speed of execution, like, ship, learn, iterate is one of them. Like, You know, you have a, like, finite time in life, anything like life, like a career, a path in a, in a company, and the number of things you can learn really depends on how many loops you can get in that. And I really want, like, everyone working with us to, like, think about it. Like, how can you learn faster from what you do, from what you ship? So, for us, it's critical. I think Elliot, my co-founder is, like, he's an amazing developer, but he's, like, the fastest builder I ever met. Like, he ships so fast, it's amazing.

AI assessment note: “Yeah, definitely. Like, actually, it's, so we have three cultural values at Photoroom.”

Q I'm going to apologize for being direct in this interview, but I'm going to apologize once and then just be direct. Some would say like, ah, it's like, it's like background removal as a service, like cool, but like it's a feature. How do you advise found, and I'm sure you've been told that respectfully, how do you advise founders who are told that it's a feature, not a product?

A Yeah, I think it happens to a lot of entrepreneurs, so it's a great question. What's important is you, I mean, to see the depth in the feature, and I think you become, like, a product and not a feature the moment you understand, like, this feature is, is giving depth to what you can do. And to give you an example, like, okay, you can say background removal, well, it's like a feature, like, you do white background, well, first it's useful for some use cases, but Like, being in the industry for, like, 10 years before that, I knew, well, yes, but it's not only that. Like, first, as a, like, a, a product, you can remove background, but you can also blur, you can, like, focus on what's important. You can do templates, YouTube thumbnail, like, uh, it's, like, actually, it opens up a lot of opportunities for other features you can build on top of that. So, I think seeing this depth that, in the product that's coming from it is really important.

AI assessment note: “What's important is you, I mean, to see the depth in the feature”

Q Can I ask a weird one? Do you want people to remove the watermark? Because it's great advertising for you, but you also want them to pay, and so it's.

A Right. In mobile subscription, like, you always have, like, there's always 10 X at least more people that are going, you know, always people that are never going to pay for it, and there are always, like, at least 10 X more people that are going to be on the free plan, so the numbers, like, it makes sense, and actually we have a lot of people that do screenshots, they do screenshots of the, of the image in the editor and just, like, remove the watermark this way, and we actually measure it, it's big, But, uh, I think it's, I mean, it's fine. Like you, people are using the product. If they use it a lot, at some point they'll pay.

AI assessment note: “I think it's, I mean, it's fine. Like you, people are using the product.”

Q I totally get you. Kamal, what worries you most when you look at the future of the space? What, what concerns you?

A I think, as you say, like, the field is moving, uh, very fast, and AI is making A lot of product that kind of obsolete. So you have to think of like, to make sure you're building like a old product with new tech. I think that's a very important part. To give you an example, the first company I made, I was trying to make like postcards and print them from my smartphone. And while looking back, the obvious answer is like, okay, that's not what you want to do. You want with the new smartphone, you want to create Instagram. Like you want to share an image of where you are and share it with your friends. That's, like, the new product with the new tech, and I think it's very important for Photorum, like, when the tech is moving that fast to, okay, what's the new way of doing that, not, like, the old way with this tech, and so you need to, yeah, destroy your product all the time.

AI assessment note: “AI is making A lot of product that kind of obsolete”

Q I do want to dive into kind of the depths of AI, because I think it's important. If we just start from the top, the one thing that I'm always questioning is like, what's more important? Is it model size, or is it data size?

A It's a question that comes, yeah, often. It's a good question. Um, I think on data, what matters is quality more than size, and on, on the model size, it's a trade-off. So it depends on what you want to do. Like, bigger is, like, higher quality, but it's also a lot slower. And what Photom does is, like, being 10 X faster than everything on market, because we are focusing on one use case, so You can take smaller model because you have, like, a specific use case, and that's very, that, that, that's key. So it's kind of a product trade-off when you talk to your customer, and again, that's one thing we do is we talk a lot to the users and understand what are the trade-offs they want to have with the AI part.

AI assessment note: “I think on data, what matters is quality more than size, and on, on”

Q Can I ask you, when you look at, you know, what you've done with your own models, given the size of your team, is that worrying for the future of AI in terms of commoditization?

A I think who benefits the most from, like, so first, like, the leader in the space, like, own the full stack. So, you know, you look at, uh, ChatGPT, they do, like, the product and the, uh, I do think you need to own the full stack if you want to deliver high-quality product. The tech informs the product, and the product informs the tech, and you build this, like, a feedback loop. What's the, um, yeah, what is it worrying for the commoditization? I don't think so, because what's important for photo room is the AI part is in the, our DNA, like Elliot, uh, my co-founder is like, uh, he wrote the book on computer vision, you know? So, and I, I also trained some models, like, uh, before I've been working on AI for the past 10 years, and any commoditization make us faster and benefits from the people knowing AI, actually.

AI assessment note: “what is it worrying for the commoditization? I don't think so”

Q So how do you do it transitionally? Because you can't say, hey, we're going to replace you.

A Yeah. No, that's why I agree with your quote. Like, it's very difficult for the incubant to say, like, okay, we're going to, like, kill the product you've been using, make it totally different, and make you useless. So that's very difficult, where I think what's very interesting for the autopilot analogy is you address people that couldn't afford that before. That's what Photom is doing, and you're, you have a small business. You can't afford, like, a freelancer to, to help you. You don't have the skills. You don't have the time. If you take the copilot analogy, you don't, yeah, if you take the copilot analogy, you don't, Have the time to learn to drive, because they have, the copilot just tells you where to go, but you still need a driver, right? And so you spend, like, 10,000 of hours, tens of thousands of hours learning to drive, and then you're good. Well, for design, people don't have tens of thousands of hours when they're starting a restaurant. They should be cooking, you know? So what we give them is more like an autopilot, where they, they come up with, like, their restaurant, where it is, and we just, like, serve that to them. Like, we serve the visual, we, we automate that part.

AI assessment note: “what's very interesting for the autopilot analogy is you address people that couldn't afford that”

Q How do you feel about the co-pilot strategy? I had someone on the show say it's the last dying breath of incumbents, which I quite liked.

A That's quite, yeah, it's, I'm quite aligned with that. It's quite brutal for them, and I'm sure some will survive, but I think the problem is where I see that being very true is if you have an audience today, like you have programmers, you have designers, well, you want to, you don't want to alienate them. You want to give them extra value, so you're kind of thinking, what can I build for them that makes their life easier, but you don't want to, like, Have something that's kind of an autopilot to replace them. That's why I take the, like, yeah, the Tesla analogy. Like, you're making planes, like, you bring a co-pilot technology, but who's buying the planes, the airline and the pilots, and you're not going to, like, sell an airplane with no, no pilot at once. They're going to, like, uh, like, go on strike and everything. So I think the autopilot...

AI assessment note: “I'm quite aligned with that. It's quite brutal for them”

Q Can I ask a weird one? Do you want people to remove the watermark? Because it's great advertising for you, but you also want them to pay, and so it's.

A Right. In mobile subscription, like, you always have, like, there's always 10 X at least more people that are going, you know, always people that are never going to pay for it, and there are always, like, at least 10 X more people that are going to be on the free plan, so the numbers, like, it makes sense, and actually we have a lot of people that do screenshots, they do screenshots of the, of the image in the editor and just, like, remove the watermark this way, and we actually measure it, it's big, But, uh, I think it's, I mean, it's fine. Like you, people are using the product. If they use it a lot, at some point they'll pay.

AI assessment note: “I think it's, I mean, it's fine. Like you, people are using the product.”

Q In the early days and today. Yeah. Because you know exactly that sellers on these three platforms make it. Product marketing is much tighter, no?

A I think you can combine both. So, we have one hundred million downloads. Like, obviously not all of them are sellers. I think the focus is saying, like, okay, I'm going to only develop, like, these features for, for this persona, like, uh, sellers or someone on Shopify, but then it's fine, like, you're not going to block other people to access, like, your, your platform. We have an API, like, people can use your API the way they want, like, and you learn a lot from it, but I think it's really important in your product development roadmap when you focus to say, okay, we're going to, like, to focus on this persona It's going to be beneficial to others, but the really, the way we build it is for this person.

AI assessment note: “I think you can combine both. So, we have one hundred million downloads.”

Q Do you think founders today are using new tech, being AI, to solve old problems?

A Well, I don't think, like, doing that exactly works, I mean, especially for new founders, work at all. I think what's happening with AI is you have, like, All past products that are building tech to serve the, their existing customers. So if you take an example, like this is the idea of the, the co-pilot, right? I mean, if you're an incubant, you're, you're a car maker, you're going to build like the navigation part, you're going to build like a, some security features, but it's just on top of what's already existing. If you're the, the, the newcomer, if you're the founder, you're building the Tesla, you're building the, the autopilot part where The input is like, I want to go to this place, I want it to be like a scenic drive, and boom, yeah, it gets it done.

AI assessment note: “I don't think, like, doing that exactly works”

Q In the early days and today. Yeah. Because you know exactly that sellers on these three platforms make it. Product marketing is much tighter, no?

A I think you can combine both. So, we have one hundred million downloads. Like, obviously not all of them are sellers. I think the focus is saying, like, okay, I'm going to only develop, like, these features for, for this persona, like, uh, sellers or someone on Shopify, but then it's fine, like, you're not going to block other people to access, like, your, your platform. We have an API, like, people can use your API the way they want, like, and you learn a lot from it, but I think it's really important in your product development roadmap when you focus to say, okay, we're going to, like, to focus on this persona It's going to be beneficial to others, but the really, the way we build it is for this person.

AI assessment note: “I think you can combine both.”

Q What do you think are the biggest misconceptions that people have?

A It's only the, the beginning, like, product, I mean, so, not optimized, like, there's so much thing people are going to build on top of what's happening right now, so just one person done, like, it's opening so much thing you can build on top of, like, it's easier to build, new ways to build, new ways to consume content, so that's, like, People that kind of judge, like, this is not quality, high quality enough, like, people have spent, like, a few months on that. Photom is three years old, so, you know, it's, uh, it's only the beginning. A more, like, practical thing, uh, I think is a second misconception is people don't like to write prompts. Like, it's kind of a command line interface, and for a lot of things, it doesn't make sense to, like, uh, to input text, I would say.

AI assessment note: “a second misconception is people don't like to write prompts”

Q And when I look at consumer subscription, I've had this out with Nico and Jason, and I do, when they're churning, 30, 35%, I'm not saying this is you, but just consumer subscription, 25 to 40% of their customers every year. How do we make that business work?

A Yeah, the dynamic is, I, for us, it's really focusing on people who are deeply engaged and deep focused. I think you need to live with the idea of, like, mobile software is, like, tenx bigger, bigger than anything you'll do on desktop anyway. So you have, like, we're addressing, like, five billion people. Like, uh, I mean, my kids, the first app they use is photography. They use PhotoRoom. You know, it's, like, one of the first software you use in your life is photo editing, and probably the last one you'll use, too. And if you want to, like, be the number one, though, you have to embrace that. So, I mean, you're addressing five billion people. It's much bigger than anyone else, any, for sure, any SaaS business. Well, people will test the app, will try it out. Then you want to focus on the one that matters the most for you. And so that's, that's, like, get the question, get the usage, get the analytics to understand, okay, this one is one we really like, and let's focus and let's make a, make it, like, marvelous for them. Let's delight them. But it's really, the top of funnel is huge, and you shouldn't, like, you should embrace that for a mobile subscription.

AI assessment note: “the top of funnel is huge, and you shouldn't, like, you should embrace that”

Q Do you think the foundational models will be commoditized, though? Like, would you be a buyer of OpenAI?

A For what we're not specialized in, like text, we, we, we, yeah, we're built on top of, I mean, that's what we do all the time. We start from, like, something that's open source, commoditized, to understand the product value, and, but then you need to, if you don't want to be, like, if you want to build a mode, you have to understand what to train and iterate for your own model. So I do think commoditization brings value to the innovation for, like, open source, but then the best, the best player will be able to, like, retrain something on top of it. And start really commoditizing. There is value in having the tech and understanding the training in that side.

AI assessment note: “I do think commoditization brings value to the innovation for, like, open source”

Q Can I ask you, when you look at, you know, what you've done with your own models, given the size of your team, is that worrying for the future of AI in terms of commoditization?

A I think who benefits the most from, like, so first, like, the leader in the space, like, own the full stack. So, you know, you look at, uh, ChatGPT, they do, like, the product and the, uh, I do think you need to own the full stack if you want to deliver high-quality product. The tech informs the product, and the product informs the tech, and you build this, like, a feedback loop. What's the, um, yeah, what is it worrying for the commoditization? I don't think so, because what's important for photo room is the AI part is in the, our DNA, like Elliot, uh, my co-founder is like, uh, he wrote the book on computer vision, you know? So, and I, I also trained some models, like, uh, before I've been working on AI for the past 10 years, and any commoditization make us faster and benefits from the people knowing AI, actually.

AI assessment note: “I don't think so, because... any commoditization make us faster and benefits”

Q Do you think the foundational models will be commoditized, though? Like, would you be a buyer of OpenAI?

A For what we're not specialized in, like text, we, we, we, yeah, we're built on top of, I mean, that's what we do all the time. We start from, like, something that's open source, commoditized, to understand the product value, and, but then you need to, if you don't want to be, like, if you want to build a mode, you have to understand what to train and iterate for your own model. So I do think commoditization brings value to the innovation for, like, open source, but then the best, the best player will be able to, like, retrain something on top of it. And start really commoditizing. There is value in having the tech and understanding the training in that side.

AI assessment note: “I do think commoditization brings value to the innovation for, like, open source”