01:28:19 · Luca Ferrari
[01:28:19] S1: Yeah, we've been using AI pretty aggressively for as long as I can remember. Certainly in 2018, I'd like to say we're using it to predict user lifetime, basically to inform our AB testing. But I would say over the past two years, with very rapid progress in LLMs we've been able to have some major breakthroughs in various areas, especially software engineering and data analysis and product design. I'll give you a couple of examples. For design, we recently actually deployed a tool we built in house called Diagram. Whether you're a designer, a product manager, a growth manager, you go to this tool, it looks a little bit like Cloud Design, just broadly speaking, but it's specialized in our particular context and fully integrated with everything else at Bending Spoons. You can just tell the tool to pull up screens for the app you're working on, so it's Evernote, for the relevant features, and then you guide it as it produces new versions of those interfaces. And it will do so by automatically following the design guidelines that the, say, the head designer for that tool has laid out in some documents somewhere. Don't have to know where they are, like the tool nodes. You just tell it what you need, and they'll give you work that the head designer would typically approve. It will automatically look into the code base to know how the different interfaces interact functionally, it will make proposals that make sense from that point of view. And then once you're happy with your proposal, it will develop the code for you, and then the lead engineer will be able to review and approve it if it's fine. Then automatically, because it's integrated with our AB testing system, you'll have a new segment where you're going to test that new onboarding flow or whatever. If you had the skills before, were a product designer, now you can do it sometimes in maybe 1% of the time. In many cases, it's actually a better result because it's so precise and in just few months we tend to miss things. But interestingly, it enables doing design work for people before who couldn't, like product managers, software engineers, growth managers. Now a lot of the inefficiencies stem from I'm a product manager, have an idea, I want to test something, but I need to wait for the product designer to be available. Then I need to explain to them what I have in mind.
01:30:37 · Luca Ferrari
I failed to explain it properly. Get three days later, I get work back and it's not what I meant. Need to explain. The inefficiencies stemming from this exchange of information, we humans are insanely inefficient at exchanging information. We're quite efficient at absorbing information. But when we have to articulate ideas, language is very... And it's better than not having language, it's very inefficient. And more so when it's with another human with whom the iteration cycle will be slow because they may not be able to do the task immediately. Even if they do, will take them time. But with the machine, we can tell it, it will do it right away, and it will take them a fraction of the time so you can actually iterate very quickly. So overall, you get to the result in a tiny fraction of the time, but interestingly, you can do it even if you can't design. So this overall makes our teams a lot more efficient. This is an example. Another example is something we call old spooner, or it stands for alter ego or alternative spooner. And it's basically an agent that lives in Slack. We use Slack for communications. That has, by design, the very same access you do as an individual in the company. So it has access to the same tools, to the same degree. If you have full access, partial access, it mimics you. It's meant to be you basically, but artificial. And you can instruct it to do pretty much anything you could do, it could do. Some things it will do better, some things it will do worse. So we have Evernote. We we have a channel on Slack where we can report feedback on things we think could be improved bugs or new features. And I was there to to provide input on something. I I was just using the tool the tool and it failed at something and I wanted to to to relay that. And I I saw live one of the best users of old Spoons by one of my colleagues. She runs Evernote and and she wrote, you know, she... In this channel, she tagged old Spoons and said, I noticed this bug. Could you please go to Moros, which is, again, back to the integration of our tools. It's it's our customer support tool that collects feedback from users to check whether it's just I got unlucky or if it's a a widespread phenomenon or issue, and then report back so we know how to prioritize it.
01:32:51 · Luca Ferrari, David Senra
And separately, can you look into the code base for root causes for this issue, and if you can find them, propose a fix, and then ping Marco's lead engineer for that particular product so that he he can review the code and push it to production if it's fine. And so she, general manager for Evernote, in maybe three minutes, essentially fixed... Identified and fixed a a bug, something that would have taken [01:33:15] S0: If this is human to human coordination Yeah. But weeks maybe. [01:33:18] S1: Forever. Like, exactly. There are many more examples. I mean, I I had to... I wanted to know the the the, you know, the the trajectory of monthly active users on on on Meetup. That's not not one of our properties recently for analysis I was doing. And and, generally, I would have to ask a data analyst, and they'd be busy. I would either interrupt them or they would get back to me couple days later. They... It would take them presumably a couple hours to give me that. I actually interacted with my old Spooner, went back and forth asking for further cuts. Okay. Just show me just for The US, just for for users on on iPhone. Just on... And, you know, I got all the answers, all the graphs in a few minutes. Perfect. [01:34:04] S0: Done. I need to go back to this because you just blew my mind. So I know you're not telling us exact numbers, but Evernote's doing probably a couple 100,000,000 or thereabouts in revenue. [01:34:13] S1: Let's say more than 100, less than 200.
01:34:13 · Luca Ferrari, David Senra
[01:34:13] S1: Let's say more than 100, less than 200. [01:34:15] S0: Okay. So there you go. That's the range of revenue. It it is profitable, and you just gave a hint as to what the operating profit percentage might look like. Right? And you're doing this to 20 people. [01:34:27] S1: Yes. That's right. Yeah. Plus the help, slightly unquantifiable help of that platform that keeps pumping out technological improvements, those automatically benefit everybody. You can allocate it by dollars in revenue, whatever. But yes, people wake up in the morning and fix bugs, Evernote, launch features, optimize monetization, that's about 20 people right now. [01:34:50] S0: So this kind of efficiency, are you seeing that in the rest of the businesses that you own as well? [01:34:54] S1: Yeah, for the most part, I think that not all functions are equally [01:34:57] S0: Are you optimizing for that?
01:34:59 · Luca Ferrari
[01:34:59] S1: No, we just try to make each business as successful as possible. It's not like we want to have like, the smallest number of people that we can. I mean, if more people create more value for customers and for Banish Spoons, we would... Assuming we can we can hire enough, you know, and fast enough, we would certainly deploy them. Sometimes we have situations where we will gonna have more people, we just don't have them. So, okay, that's a separate issue. But we don't aim to minimize the number at all, just to run these businesses as well as possible. We often find that some of these businesses, when you take them back to startup mode, so they had been large, slightly bureaucratic, sometimes political organizations for a while, things tend to grind to a halt. It's difficult to be entrepreneurial, enthusiastic, move fast, move work on what matters. If we bring them back to a much smaller size, much higher talent density, get rid of a lot of red tape. Then even though the team is smaller, perhaps precisely because the team is smaller, product development and optimization of monetization pick up again. And Evernote is a good example. If you look at the it's difficult to precisely quantify innovation, if you look at the timeline of, say, product improvements, features, before we acquired it in the, say, two, three years before and after, it's night and day. I feel very comfortable saying it's at least three times as fast under almost any frame of measurement, despite the team being much smaller. But it's really think despite is the wrong word. In many ways, it's because it's a lot smaller. So these people are against startups. Instagram was built by, I suppose, I don't know, like 10 people. I'm not sure. Something like that. I think
01:36:37 · David Senra, Luca Ferrari
[01:36:37] S0: it was like 12 when [01:36:38] S1: they got [01:36:38] S0: acquired some is crazy mind proof [01:36:40] S1: that small teams of very capable people with extreme ownership who really care can do, can outwork and outproduce vast organizations where either not enough people care or they do, but there are so many feet to tap onto and so many hurdles to overcome to get stuff done that they they fail to do so. And it's not... Mean, nobody wants that to be the case. It's more like frog in the boiling water kind of phenomenon where you keep adding teams and and and and processes and rules. And then at at some point, it's very difficult to it's very difficult to care, and it's very difficult to get stuff done. This is what