What Happens When AI Adoption Actually Works? - with Eric Porres, Chief AI Officer of Logitech

Now, the leadership team members, before there's a board meeting, they send their information first to the board advisor for the unbridled feedback. And I think part of this, it speaks to a change in workflow, which is when you embed AI into the workflow of an organization or other process, then everything follows. If you add it on at the end, then it's not as necessarily as powerful and impactful. Hi, I'm Eric Poras, the chief AI officer at Logitech. I'm super excited to be back with these two gentlemen because we were here a year ago and the few things have changed. Models have changed, harnesses have changed. AI has changed, but the work hasn't. And so I'm delighted to talk about how we eat humans and AI together to build great products and do great work at Logitech. Welcome back to the show, Eric Poras. You are one of our most listened to most favorite episodes. We're so delighted to have you here. Here's the first question, what's changed? Since our last conversation, you've been on the ground, you've been driving the transformation, you instrumented everything out the wazoo, which is the technical term. Yes. Let's change from 18 months ago, last time we had John the show. Yeah, so great question, Jeremy. And thanks for having me, and I, of course. What I would say is that the, I'm a victim of my own success. Please say more. In the sense that you poor thing. You poor thing. No, and I, as I, I'm trying to figure like, what's the best way to say this? And it's the only like the tiny little pat I put out myself is that we have now this Cambrian explosion of citizen-based AI initiatives that are happening across the company. And I think that's what happens when you get to a place in which you can go from AI curiosity to competency and fluency. So I could not give you an organization. Whereas a year ago, I could say, yeah, dabbling experiments, a little bit here, a little bit there. I can't think of a single organization within the organization now that is not building, exploring, creating something with AI as a, not just as a thinking partner, but also as a doing partner. Anything you know now that you would have, you know, seeing kind of like a thing that happened over the last 18 months that you would have done differently, or is it just excitement that people are now using it? Sure. So I probably would have done, we would have done more work on, on token calculations, of course, you know, of everyone. And this is not to say that the nice thing that we've done as a company is nobody does token maxing. Like I don't know why that even existed as a thing. Like it was such a weird epic, it only when ever with this gets published. Like in the last, you know, first half of 2026, you know, no one was actively out, they'd be like, I'm going to do as many tokens as I can. No, I'm going to come up with thoughtful ways in which I can apply AI to my work, to re-imagine workflows, to re-imagine software development, to re-imagine knowledge work in that way. So, but having said that, Heinrich, you know, with great power comes great responsibility. And like all of us, I think we could have done, again, you can't do it, you know, a priori. Well, I was going to say, would you start with that? It feels like starting with a compliance policy. It's just like, well, not really the most inspiring thing. Exactly. So not a budget, but just a formulation of an expectation. Like we knew numbers were going, right? It was what Pete Steinberger, who said, I think he said, like the growth of OpenClaw was, or in his TED Talk, he said it was a stripper poll. So, no, wasn't a stripper poll, but certainly hockey stick. And for people who don't understand, not a hockey stick, right? Not like typical exponential, his basically it, nothing, nothing, nothing, nothing, everyone in the world. Yeah, everyone in the world, maybe except me. But where the hockey stick growth, we saw the hockey stick growth coming out of 2025, in terms of growth everywhere. So adoption, usage, tokens, projects, et cetera. So again, I think we all could have been maybe, yeah, just slightly more mindful about thinking about, well, what does this actually mean? At what point on the curve, are we? And I think collectively, as a society, even though I feel like we've done an amazing job as a company, getting people into this state of AI flow, if you will, to park it back to my old professor, Mihai H. Except Mahai, we're still only scratching the surface. And so, when I think about, therefore, on a macro market basis, when I think about, what, you know, Eric Ringeoffson has said, Aaron Levy has said, and others, Jensen Hwang has said, we are only, like, the demand will continue to grow exponentially. Well, pause there, Eric, what have they said? Because I want to make sure for folks who aren't as dialed in, I know we're all listening to the same podcast, but just for somebody to go, wait a minute, hang on, what did Jensen say? You know, what did Eric Cook say, right? What would just give folks like the quick summary of what the latest thinking from those say three people is? Well, the quick summary is that in Bringeoffson's case, we're only starting to see the net positive effects of AI in terms of labor and productivity on a macroeconomic basis. Jensen Hwang will say, we should dare to do more, meaning that AI is not a great eliminator of work, but it's actually a great enabler of work and of exploration and ideation. And we have said this, you know, Hanukkah said this over a year ago, AI will only cease to be useful to us if we run out of ideas, right? So that's like, it's inherently, it's itself challenging in that way. It's like, well, are you in the idea of business or are you in the widget business? If you're in the widget business, then yeah, maybe it won't be helpful for you. And then Aaron Levy and others, who I think have a pretty sharp take on the state of the industry is that we're still very, very early. And so, and look, we see this in, you know, TSNC and chip manufacturing or the NVIDIAs below out quarter, demands greatly exceeds supply. - Here's a question for you on that. We talked to Charlie, Charles O'Reilly the other day, fellow Stanford professor about incubation. And one of the things we talked about was one of the issues with incubation and big companies is that, you know, when they get to, they get the projects of the ground, but then when they kind of need that basically the A-round, when they need to kind of go to scale it, then in many organizations, there's not an infrastructure or Amazon famous that have one, but a lot of organizations don't have any infrastructure. Question unlike the use of AI. I sense that there's a little bit of a comparison there because you see people kind of say they use AI a lot, then they say they've saved, like, end this amount of hours. And then now that the cost equation comes in, where suddenly, like, the bill to anthropics of AI is going high, then the CFO's office will go like, okay, but where do we have like the proof points? Do we see dark top line? Do we see it in our EBIT, what do we see? So there could potentially be this kind of risk that we're now seeing kind of like the cost go up and we haven't really found a way of defining the successes, like in the metric that kind of covers the cost. Sure. How do you think about all that and how do you navigate that internally? Well, all right. Wasn't it Warren Buffett who said the price is what you pay, but value is what you get. So, you know, I can't necessarily give you an exact value of, you know, me generally wearing black shirts every day. I know it makes it easier for me in the terms of decision science in the morning to put a shirt on. As Heinrich and you make it an honorary Danish person. Exactly, correct. And so your point is a good question, which is how do we measure, like, doesn't matter? Did it, did it make a difference? And one thing that we did early on a year ago was that, and I actually don't even know if we talked about it at the time. We have a project check-in tracker, simple document, right? Simple app script that I built, people can enter their information and so is your project, no matter how big or how small, we do want to know about it. And then there's a reason why we want to know about it. How big or how small is it? Does it ladder up to 2031 strategy? So, one of the things that we did in late summers, we said, hey, now that we have 2031 strategy, more or less in place, one slide summary, let's turn that into a markdown file, so that anybody who's working with a thought partner can have at the back of their mind, okay, like, again, is this tiny little project that I'm doing, maybe you don't necessarily need to measure, whether it connects to 2031 strategy, but if you're building something that more than a few of you can use in some way, create some utility for yourself that you didn't have before, then maybe you want to understand if that's connected to your strategy, great. So, the intake form has its secondary connector, which is the outtake, which is to say, hey, now that we checked in 100 and number of projects, 200 projects, take your pick, what was the output? Did it, in fact, reduce time, save cost, create value, give more time, ultimately back to in a sales and marketing organization, like our most precious commodity, the time we're spending together right now, we will never get it back. So, let's make the most of the time that we have. We all have 24 hours in a day, some of us sleep less to get more of that time, Sam Altman talks about you have polyphasic sleep so that because there's just so much opportunity in AI, and I believe it, right, like I built it, and we could talk about that, I built a, oh, we'll get there, we'll connect it to my work, to try and mitigate and mediate the amount of time I spend getting vampires into my, into my, the point is that you have to start with the end in mind. And I think that's one thing that's new for many people is if you're working at a larger organization, you're working at a larger organization. You don't necessarily come from, you know, Heinrich, your background, Jeremy, your background, even my background in it from an entrepreneurial perspective, which is say, okay, how do I go from like a V zero artifact is great. But then how do I scale that V zero artifact and what's the impact that it's going to have on others? And so that's why we instrumented this outtake process. And by all you mean there, Eric, is you're closing the loop. You're saying, hey, innovator, to go back to Heinrich's question about kind of explore exploit broadly, he's asking, how are you cultivating exploratory capacity? And what you're saying is we're having folks not only tell us about the new stuff they're doing, but then we're proactively calling it reaching back out to close the loop with them and say, did you validate your hypothesis? What did you learn? What was the impact? Is that a basic summary of what? Yeah, that's great. Jeremy, you always say it better and more succinctly than I do, which is, which is in fact that, right? So scientific method at its purest form is, okay, here's a hypothesis, validate the hypothesis, what can we learn from that now? What we did, what we've subsequently done or what we've done in that work in that time frame, is we built or a Google workspace company, and we don't just exclusively use Google, but Google workspace is pretty helpful because we created a gem, which is what others would know as a custom GPT if you're in the in the chat GPT world. And that gem is a build advisor, and the build advisor says, what are all the surfaces that you can build safely and securely within Logitech? So whether that be a workspace studio flow, whether that be an app script, whether that be something an actual agent to Gemini Enterprise, whether that be something you can do in Logitech Q, but one of parts of the knowledge of the gem goes back to the project intake and outtake form, such that Jeremy, if you're starting something that's net new to you, what might be new to you may not be new to Henry. And so now I'm already giving you a head start because now another citizen, a Heinrich, can come to, or you can say, oh, you know what, the gem will say, hey, Heinrich, it sounds like the project that you're thinking about, right? So that it doesn't interview first principles interview, help me understand what you're trying to build, what you're trying to accomplish, what are your goals, objectives, et cetera. Hey, you know what, and that sounds a lot like Heinrich's project. Yeah. Maybe there's something you can learn from Heinrich, please talk to Heinrich. So you're saying that the gem which I love, the build advisor, I think there's two levels there that are worth calling out. I just want to, you know, codify that or bookmark it. Sure. Heinrich is, it's been instructed what are kind of the toolkits in the sandbox that someone's allowed to play with. What are the ingredients that you have? You're a chef. You're going into work with the master chef, master AI chef. What are the ingredients you have that you can work with safely and confidently across across the universe, across the 7000. So it's like what are the ingredients and then what is in the fridge already to use like another metaphor, right? Like, hey, Heinrich's actually made this dish. Yeah. What are some of the recipes that have been tried across the organization to solve maybe this particular problem that you that you've identified yourself. But look, because you're in working in a 7000 person organization, different regions, different business groups, different countries, different social, folkways, mores, et cetera, how do you you won't you won't know what all 6,999 people can tell you about what is the impact of that? Like just of the build advisor and it could be anecdotal. But what are you hearing from folks who go to the build advisor for understanding of tooling available and kind of prior art? How does that then impact their go forward? Oh, my God, it's in phenomenal. So again, again, victim of my own success, quote unquote, is or victim of our collective success is now that we have, it's, it's the cold star crumble. Right. So you and you've talked both of you have talked about this many times on this show in another places, staring at the blank page. How do I want to get there? Yes, I can work with AI, but now, oh my goodness, not only can I work with AI, but I can also work with a human who has tried to solve this. To solve this problem as well. So it's almost like if we think about and and Bryce has talked about this before and it's it's Bryce Chalamel. There are more chess masters today than ever before because they have now worked with computers. And such that it's like it's the combination of the two of them are more or more impactful than one or the other working on its own. I would say the same thing applies here. And this is one of the, I think key takeaways is right like the cold star problem is no longer. Oh, let me just work with my thought partner. Well, that's, that's great. But where's the human in that other than yourself. Yeah, so like the democratization of AI truly means it's more than just, it's more than just technology. It's also people in process that that can support it. And I think that's very, very powerful because right now in large respect, right, there is a fear of well, I'm just giving my work away or I'm working with a machine. Where is the human and this also closes the loop on I think on the human experience as well. And so what I am seeing is number one. That gem is the most used gem in the company is we have company we have shared shared company gems. Number two, the projects that then get started. Start from a place of confidence because hey, I now I know what the ingredients are versus what they're not now there will always be exceptions, right. There will always be people who have either certain roles or certain experience where they want to build like above and beyond and then like, you know, pray. Pray, hey, this project will get enough energy around it that hey, we can deploy it in this way that's kind of non standard and look, there will always be non standard projects. There's a bell curve of projects, right. Solving for the bell curve, though, gives the bell curve a much higher slope. And I think that's what we are right now as as a company. Now I was curious a little bit on if you do you think that there is kind of like these plateaus that you hit where humans have to kind of catch up. Or do you think that the experiential curve that we see in the eye world kind of outside the organizations that we work in is kind of, yeah, yeah, they will see the same curve internally. Like what if you look at the next six to 12 months, is it just more. Well, yes, so I think if you were to zoom in and it was funny what is as you're saying that hammer because I was thinking about that before, which is if you zoom in on the curve, what you actually probably see are these little. These little plateaus now ordinarily in in in the like a technology timeline, you would see those over a longer period of time, right. Whether that be the gardener hype cycle, you know, peak effect, you know, trough disillusionment and then and then moving on. I don't think we're necessarily in the I don't think we're in the trough of disillusionment phase anymore. But I do think there are these slow and steady peak plateau peak plateau peak plateau. So what do I mean by that? What does that mean in practical terms. It means that either the next generation of model X has become sufficiently capable such that the again, using the term from either malloc, the jagged frontier has, I think, become less slightly less jagged. So there's more of a curve than there are spikes. It's like analog versus digital, I suppose in that way, which is weird because analog humans digital machines. But the. So yes, there are plateaus because when something new comes, you're like, well, shit, what did I not understand or what did I think the previous model wasn't capable of that it is now. But that also means then that you have a responsibility as a builder to think about like to not be so wed to this idea that this thing that you built six months ago, because the technology wasn't there yet. It's suddenly there and now you have to let go that you have to be able to let go. And I don't necessarily agree or disagree with Jack Dorsey and a lot of things, but I want to say 10 years ago he did an interview in the Twitter days, where he's like, yeah, I've said no, I've had to let things go things to fall on the cutting floor a lot and I have to be okay with that. You know, this idea of saying or, you know, subtract lighty clots, I have to say no to more things than I say yes to. And so what that means where I'm going with this Heinrich is, and I've written about this a lot in the last six months is this notion of, you know, skills as software and abstracting the capability that you're building into a layer that any agent in the future any harness any control plane in the future can really understand and take advantage of. So to a certain degree like the the PRD or the PRD that you're writing and building over time becomes even more important because now you can give you can give that document rather than writing a piece of software to do the thing you want to do if you can give that document in that set of instructions to this capable AI partner that you have that is able to execute it for you in ways that you were able to do it before. Do you have other of those concrete kind of suggestions? So I mean, obviously, skills of software is like a very concrete thing like what are the things that are going on internally, that you kind of go like this is what other people who is not as far down the slope with I am should do. All cases that we see is just works very well. Yeah, no, good, great, great question. So, so the buy and leadership buy in, and I mean, leadership from the top Hanukkah, right? So leadership buy in is especially in an enterprise organization, super important. What do I mean by that? Every global huddle that we have. have as a company, there is an AI in action moment. So every month, there's a different group of different organizations. So it's not representing, it's not IT, it's not marketing, it's everybody has an opportunity to contribute to the AI at action moment at the global hub. You were saying this is codified. This is a priority. Whenever we globally meet, we spotlight something about AI. We spotlight something about AI and that AI contribution can be something that you might think of as as panatic, like okay supply chain logistics organization, but okay, nevertheless, okay, let's actually shine a spotlight on something which most people might think of as that's kind of boring and uninteresting to me. So we're not necessarily looking at all of the stars and dogs, so to speak, there are some stars, there are some dogs which we then promote and become stars of the show. It's also at the leadership team level, so every week, the leadership team has an AI in action moments within the leadership team as well. So it's not 18 people going around the table for 18 minutes, it's more like one person or two people sharing something that they're doing either personally professionally for themselves or representing someone who will come in and present to the leadership team. Here's something that we're doing in gaming or here's something that we're doing. Do you facilitate that? Does Hanukkah? Oh, Hanukkah, Hanukkah does. It became, we talked about and I said, I think this is really, this is super important and she's like, yeah, I agree. And so now it's so it's codified at that level. Cautification and we didn't have this when we talked last time, another gem that we built is because Hanukkah talked about this at the women's CEO 50 events as fortune in October was, you know, she talked about having an AI board member and what does that mean? So what do we have? Our parallel to that is we've built, and I worked with Sam, our head of legal and also J had a digital office and Hanukkah as well to create a gem, a board advisor gem. And the board advisor gem, by the way, is available to anybody in the company. And so now the board advisor tries to play the role of, could be curmudgeonly board advisor, it can use a secratic method. It's a reasonably, I would say, reasonably advanced gem that we iterated on over the course of, you know, six weeks before we made it available in December. And now the leadership team members before they go to the board, before there's a board meeting, they send their information first to the board advisor for the unbridled feedback. And I think part of this to Jeremy, it speaks to a change in workflow, which is when you make, when you embed AI into the workflow of an organization or of a process, then everything follows. And this is the same thing with training, right? If you're like, okay, well, I need to get trained in AI. Fine. What is it going to do for me? But if the leader, if the person that you're working with says, hey, you know what? As part of our embedded process, we are going to make it part of our process where we have to work with AI in some way, shape, or form, then everything follows from from that. So members of the leadership team, in that case, then are required. It's not just that this board advisor gem is available. It's actually a baked into the workflow. Prior to you submitting your slides for the quarterly board readout, you must get feedback from the gem board advisor. From the board advisor. Yeah. From board set, we call it, we call it board sets. Now, again, I, I won't speak to, as I can't, and I, I won't speak to the mechanical codification of that as like the does every single person as part of the leadership team do that 100% of the time all the time. No, but it's an expectation. So right, it's, it's like, that's the expectation. And, and what I find too, when I, when I use, I always learn something from the board gem, I always learn something from, we have a presentation coach, gem, the other most popular gem that we have is a presentation coach where I've inserted lots of different great presenters, whether it be, you know, the the cynic, why, how, what, whether it be, you know, Spielberg, thinking about ways in which we present ourselves, right, because as, as much as we have technology, what are we were storytellers? We, we have to create a compelling narrative. And when you're, when you're trying to create compelling narrative in slides, like that is inherently a mismatch in terms of like, when you're sitting around a campfire, are you going to present a slide? No, you're going to tell a story. And so when we have to create presentations, we often lose some of the humanity that's associated with that. And so ironically, a presentation coach gem brings some of the humanity back into the story that you're, that you're trying to convey. So, so, Heinrich, that goes back to your question. So the codification of embedding AI into workflow and business practices is one, the board sense gem is another, the having a champions, a very public champions group of 175 people across the world is another one, because not only, again, for as much as I think we can kind of, you know, catch and catch and release using the AI builder gem, right? That only that only gets you so far, right? That'll that may, maybe that catches 50% of ideas. The other 50% or it's the watermaker and you know, half of my ideas are good. I don't know which half. The other 50% is then caught and and extended through having a very public and active champions network. I want to get to deletions because that's something I know that you've been thinking about. And so that's kind of just earmarked. That's the next thing. Would you drill into champions networks just a little bit more practically? How do you cultivate that? It seems like that's kind of a community of practice who's deputized to spark ideas and availability as a resource in the organization. Correct me if I'm wrong. But can you say just a couple more lines about how does a champions network work before we get to deletion? Because I know that's another big principle. Yep. Yep. So champions network is an entirely volunteer organization. I started it in June of June of last year. It went in then it's it's grown. And we have some people that that matriculate in and out. No one necessary. Well, people sign up for extra work in some ways. And the extra work is, hey, if I am in the legal organization, I have there are two people or now, now they're more and now there are five people, but there are two people in the organization who started as champions saying, hey, we're going to actually team up together because there are different parts of the organization in terms of what they have to do from a law perspective. And how do I take this embedded knowledge that I have built over the end number of months and then share it with rest of organization? There are other champions that are because we have physical offices in different places around the world. There are other champions who are site leaders, if you will, for within in in cork as an example. There are recently big office in in cork, Ireland. They're the ones who are running the local huddles. And there's an AI and action moment that happens within the physical site sites. And so when that happens, then you have champions who are part of the the steering committee of ways in which to bring capabilities in and interesting projects that are taking place in in a physical location and representing it to each other in addition to running the weekly office hours as an example, right? Hey, I'm here. I'm working on and I still do it today. In fact, as soon as we finish this session, my next session is an hour where it's an open zoom. And anybody can go out with a question or anybody can show up with a question. Sometimes two people, three people, five people, sometimes no people show up. And that's okay because people are busy, which I get. And so but that hour for me is a dedicated hour. All I'm thinking about, all I'm going to work on something that is wholly AI enablements and and related. So what I'll probably do if nobody shows up in that hour coming up is I will work on, okay, here's some things I've been thinking about. Now I want to share it with this. We have so we have a private group chat of 175 people. And those champions also, by the way, you also have like what are the why do you show up to be a champion? Well, you get certain perks. Like maybe you get access to models that become available before others. Maybe you get access to new capabilities that we're working on that are still in development within our own control plane harness that we built for ourselves. Maybe you get access to MCPs that are, you know, governed and secure and capable that we want to push out to those ahead of time. So, you know, champ champions are are they get the first look at one of the perks is getting first look at technology. And another perk, of course, is you can just go back to from the top. A couple of champions of corner have lunch or dinner with Hanukkah, depending on where she is in the work. That's cool. Erica, let's pivot a little bit over to I know that you have like a pretty robust personal kind of like set of tools. And so do you mind just giving people the latest on what is your current setup on a purely personal level? Like do you have stuff that you just run on your own computers and do you have a different between that and your work computer and like what is your what is your stack look like? Sure. So my stack right now, I've got two screens open. You're you're one. My second screen is my I think we talked about this the last time my five open tabs are still JGBT, Quad, Gemini, Logic Q, Notebook. Those are the five places. And and in my my other second screen here, I have have co-work. So I've been a big user of co-work since it came out in January. And I would say that for me, the blends, because all I think about thinking do and create is something related to AI, I have certainly spent a good, a healthy portion of my time in co-work and inventing ways. I did not feel comfortable with my own capabilities to truly sandbox OpenClaw when it first started. And so I said, okay, well, what are the alternatives for me? And I said, well, I feel good enough about you anthropic ins practices and systems, et cetera, that I'm going to take the leap and really dive into co-work. And so within co-work, not only are there the regular connectors, but as you were describing Heinrich, I also built my own MCPs that run on this very computer here that I can connect to anywhere around the world. And those consist of their really five MCPs that I created. One is for WhatsApp, as I'm sure many your listeners as well as myself. And this is also publicly available in my GitHub repo. Such a difficult MCP to get to work, right? Like you kind of have to re-auth it a bunch. Yeah, you have to re-auth well, although the authorization now is actually much easier with the Baleys. Anyway, the point is that we are all part of many communities of practitioners, thinkers, learners, doers in some way, shape, or form. I do not have the time to read all of them. So what I've done is I've created a skill that codifies what I'm interested in and what I'm not interested in. And then I apply that skill to a daily summary of the conversations that are happening in six different WhatsApp groups of different AI practitioners. There's some overlap with some, but there's some that are unique. So what do I want to learn that is unique to each group, a practitioner group versus one that's more macroeconomic theory, et cetera. And then also what's the what's the crossover? So that one has been super, super helpful. The next one, and Jeremy you talked about it before, is whoop. So I've been a woop member since 2020. And what I found earlier this year with the arrival of co-work is that I was spending so much time, I became Simon Wilson. I was bitten by the vampires. As I know, do one click into that for folks who aren't familiar with the because that was going to be where we go next and you just went there. So for folks who aren't familiar with the concept of the AI vampire, would you just kind of describe the concept and then talk about your own victimhood? Yeah, OSHA. Well, so like the great thing about about working with AI in this way, it's like it's the perfect slot machine in terms of variable intermittent reward. So you could be working on you have you have an idea that you want to bring to life. This is this what's that problem? And you build something you're like, oh wow, this is amazing. It works. They're like crap and as high as the authentication was crap. So then you go back into it and you work a little bit longer like, oh wait, but I know it's going to work this time. Let me deploy that piece of code again. Let me test it myself. And then you know, sooner or later, you're like, oh wow, it's four in the morning. And because the variable intermittent reward of reward cycle of building in this way, build tests, iterate, refine, something broke. Okay, now, oh, now that I just built this, oh, here's another idea. And so like your idea, idea flow can almost happen in perpetuity. Hence the polyphasic sleep and the other things. So and they call that being bitten by the AI vampires. They call that being your life without you kind of being aware of it. It drains your life and it's this sense of productivity, discovery, et cetera. And what basically gets sapped is your sleep or your right? Is that in my sleep relationships, health taking you to take your pick. So. And this is a well documented phenomenon. There's a well documented phenomenon. Right. Yeah. I had that the other day when my wife came into bed and I was sitting there talking, you know, through my to my bot, conversing and she was like, you really rather would like to talk my main bot is called EG. And so you would rather talk to EG right now that talk to me. He's like, kind of and you're like, okay, this is a wrong. No, Henry's, Henry's famous line is not not that. Okay. So Eric, what have you done? What have you done? You know, configured to protect yourself from the vampire. It sounds like I'm hoping you have a solution for us. Well, I do have a sense of freedom. And so in my case, because I've also measured myself in some, you know, part of the quantified self movement for, you know, over a decade, what I noticed is that, you know, my my sleep, you know, unsurprisingly was was suffering my sleep quality sleep score. I know, for me, it's one of the most important like one of the things, right? Like nutrition, morning sunlight, sleep, and regular sleep. So, you know, no matter like going to bed roughly at the same time, getting up roughly at the same time, that was not happened. So I said, well, what can I do to change that? So who has an API? And so I said, well, I can take I can turn this API into an MCP is if I turn it into an MCP, then I can use co work to converse with it or I can I can fetch that information. By the way, co work also has a I message send receive capability. So in the evening, I can look at my overall strain. It will look at my overall strain and capacity score from the last 24 hours and say, hey, Eric, you know what? Probably time to put the pens down or close computer, relax or don't do it, don't do anything. So that's the evening, that's the evening check in. If my score in for those of you who either use Cooper or others, right, if I'm in the kind of yellow or the red zone, then in the morning, so I have a set of alerts and summarizations that take place in the morning for me as well. And it will modulate the amount of information I receive first thing in the morning on the basis of my recovery. So if my recovery is crap, I'm not going to get that same kind of feed that I would get versus if my recovery is in the green, if you will. And you're saying you're saying AI is enabled you to summarize and to access information that would either to be unimaginable. And your default is, I want to consume it all, like somebody at a buffet, I'm just going to eat everything. And what you're saying is you now have a data backed governor on how much information you get fed, which is based on your recovery. And if your recovery is poor, the AI goes, buddy, the buffet is closed today. Yeah. It's kind of interesting because I think we obviously were all reading the habit books, you know, like back five years ago, right? Like the best way. And it is interesting how I think many of us who's very deep on AI now is not trusting ourselves to build a habit, but we're trying to get the agents that we work with to basically not just to what's it, right? It's the same thing, like my instinct, like every time after 10, 30 that I say, hey, we should start working on that. It goes like, yeah, but maybe we should wait until tomorrow. Maybe you should really should end. And so Jeremy, right? The ultimate proof is that it's in the proof, which is right, whereas before my sleep score, my recovery score was trending into the like 60s. Now my sleep score recovery is trending into the 80s. You have quantitative evidence that this is working. I have quantitative evidence that this is working. And I add, by the way, I'm reading like more physical books now and doing things in the evening other than what I would be doing. For folks who are watching the video, Eric could just held up a copy of Lighty Clots' Subtract as an allusion to a previous guest that we had. It's a great book. You should read it. Eric, I'm still dying to get to deletions, but I, but you've like tempted me down this rabbit trail. How many others at Logitech have you assessed are suffering from the AI vampire? And is your MCP-powered, whoop-powered quantitative solution helping them as well? Like I just want, if you think of yourself as a lead user, so to speak, are there a lot of other people getting bitten by the vampire? And are they now asking you for intervention support as well? Well, so, you know, prior to this role, I had, I had run an innovation software group within Logitech as I had sold my company in the Logitech about needing performance and needing hygiene, et cetera. And what I led was something called Smart Habits, which is still available. It's the software smarthabits.logitech.com. And Smart Habits was all about finding the space between the space, something you hear a lot about in jazz, which is, hey, if you're working for a certain period of time, you're probably not hydrating, or you're probably not doing any deep breath work, or, you know, take a break from here, or, you know, what, and stare off into the distance for 20 seconds, you know, 20 meters for 20 seconds so that you don't have eye strain, right? Most people in the world, in the United States specifically, 75% people are chronically dehydrated. So like what Smart Habits did was it would snudge you to encourage you to hydrate more in the morning versus the afternoon. This light that that works behind me right now, this light is set to the circadian clock of this area of the US where I live. And so the light will change its intensity, as well as its color throughout the day, because you want as much light in your eyes as you can in the morning, and then you will gradually want to reduce the color and intensity during the evening. So this notion of this nominal notion of habit creation and habit enforcement is something that's been deeply embedded in me. But even then, again, in spite of that, right, like I still needed to instrument, I still either way to not only instrument intervention, but then also have a way to measure it. And ultimately thinking about, you know, adding, you know, adding years to your life and life to your years, like I want to add life to my years and, and sleep as we know from science of the last, you know, century, especially the last 20 years is one of the most important ways to do that. Can we get back to, you had two MCPs you mentioned, you said you have five that you use. Sure. Okay, we get back to the other ones said, "The boss, there's about to copy everything that you're doing." Right. We talked about what's up. We talked about LinkedIn. Believe it or not. There is a way in which you can create a LinkedIn MCP such that so because, right, like I'm, there's certain things I'm interested in. There's certain things I'm holding on. Interesting. LinkedIn is recently announced right there. They're getting, or at least they're, they're deprioritizing with 94% accuracy, like the amount of AI slot commentary that shows up in LinkedIn posts. I spend very, very little time in LinkedIn now than I did before because, again, similar to WhatsApp, I have a skill. Here's what I'm interested in. This is the kind of content that is relevant to me in my role. The people I follow, the people I pay attention to and please summarize that for me, rather than me having a hunt pack click my way through, "Okay, I want to find what Jeremy's saying. I want to find what I'm saying. I want to find what Ethan Molick's saying. I want to find out what Bryce Shalamol is saying. I want to find out what Greg Shubba's saying. Now I can have that, you know, summarized for me in a much more intelligent way. The most important one, Heinrich, for me, is a, and it's not just an MCP, but it's also, is that there's an entire system and it's like, it's a whole session of its own. I wrote about it. People can find it. The blank page is not deep work. I built my own memory palace of every conversation I've ever had, every coding session I've ever done with AI over the last three years is now retrievable via either MCP or direct call to a super-based instance, which is now has voyage. So voyage is a company that's now owned by Mongo. Voyage does embeddings and foropic actually recommends voyage for its embedding. So I've embedded my knowledge to help me in the future, every new session I start, almost every new session I start in a certain project. I say help me use use deep memory, which is what I call it, to help me build context about some of the things that I've been thinking about in this topic. It's like, remind me what I think, what I think about this, or remind me what I think about this. Yeah, exactly. Yeah, because, and because I can, because it also has some pretty intelligent re-ranking that you can do on, on the moment of, of demand, the chunking, the embedding and the retrieval now become part of my, it's just, it's part of my workflow. I cannot begin to describe in the time that we have left, how valuable this has been for me, like, imagine every conversation you had across all of the surfaces. So whether that be Gemini or GPT or Claude, co-work, coding sessions, having that all retrievable at your fingertips. Give us one discrete example of it impacting you in a meaningful way. Just so, because I think that will really help people imagine the possibilities. Just one time you go, wow, deep memory, totally changed the trajectory of this piece of work or undertaking. Yeah, both so, so most of the presentations I give, when it comes to presentation building, like, yeah, how do you build a presentation while you need to build it with context? What is your context? Is it personal? Is it professional? What are the concepts that you have explored, thought about, tried and failed, or better to try and fail than failed to try? And so I will say, hey, I am thinking about creating a presentation for the following. Use deep memory to hear some hypotheses I have in this, for this particular presentation or an article for writing. I want to go back first and think about, like, have I covered this, what have I covered before, what have I thought part of the way through? Have I built something and broken something, and have I invalidated a certain hypothesis that I have, because, again, because like, the role that I have is one where it is truly horizontal in nature, that it is hard to sometimes keep track. It is legitimately hard to keep track of all the different projects and work that come through my brain. And so I want to use this as a way to ground, to provide some kind of pivot point for my own thought, because it's like, okay, I've now had this full-crime, now I want to be able to move the world in this direction based on this full-crime, and this full-crime is based on the building, my own context, how I've thought about an idea or a topic or project or an also time-bounded to Jeremy, right? Because something I thought about two years ago may not be relevant today. And so that's where you can do re-ranking your retrieval, and there's some interesting ways in which you can build re-ranking of your own thoughts. You know it's interesting, using a timer, I think. From a cognitive, so one kind of age-old tactic of innovators, scientists, etc., is keeping a commonplace book, you know, of favorite quotes, favorite anecdotes, etc., it could be a spark file, different people have different names for it, but the only way it's valuable, like this is just like a fun, kind of anecdote, not to mansplain neuroscience, but the only way it's valuable is if you read that, right? So Steven Johnson, who we have on the show, is one of my, you know, heroes of authors. He has a three-monthly habit if he always rereads his spark file. And it's, you know, it's hundreds of pages long, but the point is there's this idea that Jim March, he mentioned that there's something to what he calls the "simultaneity of arrivals" that it's when he calls that the garbage can theory of innovation. Primarily, innovation consists of the "simultaneity" of ideas arriving, and if you think about your brain as kind of hurtling through time and space, an idea that you had two years ago is, is completely relevant for different reasons today than it was a year ago, right? And now with deep memory, I think it's kind of cool that it basically effectively, it expands your surface area for the "simultaneity" of arrivals, which is a very nerdy sounding statement, but I think you know it. It's nerdy, but lovable, Jeremy, and you're right, because the other thing I have here, you don't see. But lovable. I have, I have this, I have this whiteboard, and I could not work, people who work, who work with me over the last 20 plus years, know that Eric does not work anywhere without a whiteboard. And so the whiteboard is a place where I start to piece together some concepts first, right? So the concepting is blank page, but it's blank whiteboard in this case. And it can be little points of light. And then I say, okay, now that I have this, sometimes what I'll do, Jeremy, is the other thing I've done with deep memory, is I've now created the ability to create an embedding. I take a picture of this, I say, remember this, it immediately posts it into deep memory. And now I say, okay, now that you've remembered this point in time, now I want to go explore these vectors right now, and I don't want to, oh, let me go back, oh, what was that thing I did? Oh, what was that project that was? Oh, crap. And then you're spending all that time going through retrieval, when instead I have the retrieval instantaneously. Dude, you just gave me such a cool idea for my own life, which I'll tell you offline, but thank you. It's right here. Okay, before we wrap, we have to hear about deletions, just because I've said it now five times. So, and the context I want to give here for the listener is Eric, and I'm going to give you a compliment. So plug your ears so that your head doesn't get too big. I think Eric has done probably a better job of instrumenting AI adoption in his organization than anybody I know. And there are lots of really cool things that Eric could show us. If we had two or three hours, you could show us, you know, by person in the organization, their proficiency in all the four ways. And yet, you've written quite eloquently recently, Eric, about there's a better measure than maybe typical metric dashboards are measuring. And you talk about deletions and also creation. Can you talk about why that's so important and how leaders and maybe call it transformation architects should be thinking about what metrics they really should be tracking? I feel like we'd be in this reading, give you a chance to talk about this. Well, no, it's a good question. And I don't think it's certainly not perfected in my own mind or in my own practices. What I will say, and we talked about this earlier, this notion of dorsi saying no, and leaving a bunch of things on the category floor, is like, it's really getting to deletions about getting to the essence of the thing. What are the core, if I go back and think about my martial arts practice of 20 plus years, there are a certain set of routines, motions that one can go through, whereas if you understand those motions, those basic forms, it is the building block for everything. Piano, I play piano since I was four years old. Scales, practicing scales, as people talk about it, practice makes perfect, but it also makes permanent. Scales are the building block. So when you start adding things on top, right, layers upon layers upon layers upon layers, you can dilute the essence of the thing you were really trying to master. And so deletion is about being able to say no to ham-hawked practices because that's the way we've always done thinking here, dung things here before. And so the encouragement that I've been giving certainly to the champions, the champions are the front line of this, or they get the brunt of Eric's ideas around this, is like, your job six months ago, I said, your job is to no longer come up with an idea, but actually come with an artifact. Come with something because the ideas are now cheap, execution is everything in distribution, I suppose. King, we can talk about the later concept. So come with an artifact, but when you come with that artifact, tell us that the artifact that no longer exists. Right, so when you bring a new thing, you have to tell us what you deleted. What did you delete? Exactly. What did you get rid of? Was there an old report? Was there an old, like, we, as a species, and I know this from being able to see it, we work with an unnamed dashboarding company or data visualization company. As a company, we have have thousands of these visualizations, thousands upon thousands of them. The actual number of ones that are used are more like four percent, four or three percent of the thousands actually get used repeatedly in one way. So why do we still have all of these dashboards? Because it's what we've been, well, that's what my KPI has been. That's what I've been trained to do. So I have to do more of that. And I think if we look at examples in history, most recent history, again, for all the challenges of Elon Musk, right, like, well, why can't we land the booster rockets back on Earth? Why can't we? Well, it's always been done where we've jettison them off into an area of the Atlantic or the Pacific that then Bezos has gone down with over a 30-day period and like, you know, lifted up the Apollo 11, you know, rockets that are now in the Smithsonian, great. That's something that we did for decades until somebody said, well, no, that's stupid. We don't have to do that. We can recover what we, what we have. So deleting the process, deleting the jettison these rockets is the way in which he said, now I can actually create something new. I can create this repeatable new process, which will change space exploration forever. So this is a real world example. So when you think about this, I love it. When you think about this as a measurement, how do you measure deletions? We talked with Lighty about how one of the challenges with subtraction is there's no evidence of the subtraction, right? When you add, you go, look at what I added, right? When you take away, people go, there was nothing there before, right? And so, and so when the work of subtracting is actually a sophisticated, elegant, nuanced work of innovation, how do you then log it on the dashboard? How do you give somebody credit for a deletion? It's still a work of progress. I wish I had a very clear answer for you. I would say this to your point is like, if we meet together in six months' time, this is something that I'm very focused on right now, which is this very problem, because there are, and I actually, you know, Lighty, when I wrote about it, he reached out to me and he's like, hey, I'd love to talk to you more about this problem because it is a problem that is not sufficiently addressed by current instrumentation. Do you think one of the issues with us now dumping everything into databases is that when I went on a sabbatical, I had these small notebooks, and at the end of the notebook, I basically put the notebook down, and I had to, from memory, percolate anything that I remember for then into the next page, and then by the time I did the sabbatical, I had this notebook with basically what was the best thoughts by using the method of basically what could I recall, right? Now you have now this database of a way you dump like all your clock code and go work and all this stuff, it kind of like puts in there, and in many ways, we're now asking AI and chunking and vector databases to do that percolating for us. So I wonder if you've, what you're thinking in making sure that what we can recall has a filter, and the filter is something that is deeply human to us, so that when we then have a recall, the thing that we remember is not just anything that we ever thought, you know, set a done, but it's the thing that we felt was important when we did it. Like, almost how do you put a weight on retrieval, so there's this notion of re-ranking, and re-ranking can be time-ordinal boxed, re-ranking can be done by what I described before in terms of ingesting, like here's what I'm actually most interested in right now, it starts as a whiteboard, then it becomes an embedded retrieval object, and then the re-ranking that comes is the similarity to that re-ranking. At a separate time, we can also talk about, and I'll probably write about this over the summer, is auto-research. So, you know, Carpathy published auto-research as a thing to do with code or be able to do loops, and so I've now taken that, and Azim from exponential view also has done this, which is created an auto-research hypothesis, hypothesis test, iteratively steel man argument does this hold water, does this hold weight as well. So I auto-research my own hypotheses as well, so it's a combination of retrieval, temporal, indexing, and auto-research to come up with final, like, what is the actual argument? Because I can be right, like, I can be wrong, too. I'm happy to be wrong, and I'm happy to be proven through the scientific method. Here's why you're wrong, as opposed to just it being a set of opinions that I have based on my own objective, my own objective reality. So that's the answer to the question, and I would say, it has been net more positive than it is negative, which is why I keep using this method for myself in terms of my own a smarter way to have a maintain a memory palace based on the work that I do. I think that's a perfect place to end. This week we could, as is clear, we could talk to you for hours, and we can just do like a weeklies. Why don't we have a weekly office hours where we just talk about this stuff, right? This could be the weekly BTP, like, the subscriber-only baggage. Okay, Jeremy, he's such cool and so nice to get him back on. It's very inspiring always. We also, as people might not know, but we don't know what subgroup together. And I am always very excited when Eric kind of like drops another GitHub MCP. That's right. You love the GitHub. You love the GitHub MCP. I mean, I think it's because I'm a non-engineer and then suddenly like all this code that's been available on GitHub through all these years is available to me. So it's like a little bit like somebody who finds like a library for the first time, but just things I can use, right? And so I'm very giddy about GitHub. GitHub? GitHub, giddy up. Do you want to go? Do you want to go? Should I go? Should I go? I mean, I would say he's so knowledgeable. He's got so much experience, not only building from first principles himself, but also enabling others to build that there's something for everybody to live in this episode. I thought, you know, if I just think about some themes to see out to me, AI and action moments feel like there are many layers in the organization where that matters. In the all hands, Hanukkah is highlighting AI and action across different functions. In her senior leadership team meeting, they're highlighting AI and action moments. Then he mentioned even at different offices at Quark, Ireland, right? There's AI and action moments. To me, it's a very kind of portable tactic to say, anytime we're gathering folks regularly, that spotlight how people are leveling up, augmenting, amplifying themselves with AI, that's, and it's just, anytime there's a gathering, how are we spotlighting AI augmentations? Me, it felt like a very, it seems that's authentic to what he's done there, and I love that as a simple tactic. I'm always looking for those kind of simple, steelable tactics. I also think that he seems to be one of the few people that I've seen that manage to be very hands-on, but then kind of pass that hands-onness onto the organization. You know, he seems very accessible, obviously, you seem to have office hours weekly where people could just call in and do it. But also, I like this idea of using gems as an object that you can then have people rally around. And so I think a lot of AI development is happening now in the different parts of the organization. Doing something that is kind of more generic and available to everybody seems interesting. And what it reminds me of, and I don't know if somebody mentioned in one of the podcasts we have that there could be this moment in time where the HR officer, which now now would be not just human, but also agent officer, would be a chief resource officer. And the chief resource officer would somebody who would help the organization become more efficient and better of growing the organization. I think Moderna actually did that they merged HR and IT. I don't know if they've called it a chief resource officer. I do like that. I think there's something there. And when so, when Eric is talking to him, I was like, "Ah, maybe this is actually, I know he's chief AI officer, but in many ways he is becoming the chief resource officer, like the one who are creating resources for other people at Logitech to then use to become more resourceful of themself, right?" I think it's like a such an interesting kind of way of thinking about it. Yeah, I mean, it fits your framework, right, of resourcefulness. And I actually love that. I think there's very little difference between helping someone get augmented with AI and helping someone become more entrepreneurial. There's a high degree of overlap between those things in terms of agency, in terms of problem orientation, in terms of bias towards action. And I think that that gets captured actually in chief resource officer in a really nice way. What's nice about it is when you hear about the different gems, obviously in many organizations, you feel that there's still a bit of AI theater going on where somebody have done a few things. And then people talk the whole of that and basically says, "Oh, look, we're doing AI stuff." And you get the sense that, yeah, it's like a little thing over in the corner. He seems to be just by a dacity, having kind of like this abundance of stuff that he's making available to the organization and helping the organization develop himself. And what's nice is that when you can go all the way from kind of super small kind of like features all the way up to the board level and obviously logic being a listed company, then I feel you're really shown that you can kind of move at all levels of the organizational stack. And that I think is actually is fairly uncommon. Well, it's a very unique personality for someone to be able, and I've heard this said before so this isn't it. original idea, but you're just from you're promising it for me. You've gone beyond the prompt Henry. But there's something about the person who can have a board level conversation, a CEO level conversation, and a frontline worker conversation is a unique persona. Not everyone is capable of doing that. And if you think about Eric is kind of a prototypical chief AI officer, you need somewhat, I mean, he, I know, I don't think I'm speaking out of school to say he's in regular communication with the CEO. And he's got a weekly Zoom where anyone in the company can log on and talk to him, you know? And there's something about being fluent in that kind of spectrum of called hierarchy or seniority that's really necessary for a chief AI officer or chief resource officer. Then the last thing was this kind of idea of cause of adding more specific MCPs to your world, where you create a database that's accessible to your foundation or model of choice, which make me, I'm going to offer this conversation, make an MCP version up beyond the prompt, so anybody will be able to get MCP access to all our conversations. And I'll see how that go. But for me, for sure, it'll be usable. But I do think that we are probably getting to the point where a lot of us have made different bots, a lot of us made different skills. What we have not necessarily done is to figure out how do we architecture the different MCPs, kind of the servers that we need to create that has what data and how do we get it in there and all those different things. And so that might be something that I'll do over the summer. To me, one thing, actually, that it's a non-trivial thing. Sometimes the trigger is sometimes implicit. For example, like a deep memory. The trigger for that is Eric's going, I got to make a presentation, what have I thought about this? Right? Or I'm writing an article, what have I said about this? So sometimes the trigger is kind of obvious and it's on the user. But then the other thing, Henrik, that I notice is sometimes the trigger has to be, you've got to build the MCPs so that the loop poses itself. So for example, his whoop MCP is connected to his daily briefing MCPs, right? And so it's not incumbent upon him as the user to trigger that connection every day. He has actually instructed his briefing agents to consult his whoop MCP, his whoop data via MCP before his briefing agents determine how much to give him, right? Which I think is pretty interesting. And I think a lot of us are now trying to figure out what do we do with memory across our different agents. I'm a little bit, I post the question, I don't think I posted it in a way where my point was kind of like, probably made, but I do worry that we have to be careful about how we position our memory stack. And what I mean by that is Eric the other day sent around to a few of us a script that he created that would allow you to basically create a skill file based on on how you like to write so that you could write more authentically like you. Now, the question then becomes quickly, does this have to be a skill file about how you actually write? Or should this be a like, how would you like to write? Right? Because that's the difference, right? Or is this is when you write your finest stuff? Or is this when you write to people? And when you write it to people, it gets more human. And so I think there's all these nuances that we'll have to kind of grapple with when it comes to how do we store the digital versions of the stuff that we want to retrieve about ourselves and be a little bit purposeful about that. And I'm not quite clear yet that I have a perfect idea of how to do that. Yeah, I'm trying to see if there's anything else. I love that just to recap, you know, stuff that maybe folks heard, but we'll appreciate being reminded of. The most used gem in the company is the gem builder advisor that not only lets people know what are the tools available to them, but also what's the prior art? What have others tried to facilitate human connections? I thought that was very cool. I like that too. And then of course, the second most used gem is the presentation storytelling gem. And I love what he said, what's the purpose there to prevent folks from losing their humanity when they're telling stories in the company. I thought that was beautiful. Anyway, again, something for everybody here, I'm excited to, I think when he comes back, he owes us more thoughts around how to actually measure deletions. That's one thing. And then the second thing was it was right at the end. What was the second thing he owes us? He said, we need to talk about it in six months. I've deleted that. How about that is the code word? The code word is the answer to the second thing. If somebody knows it, we'll send you a book, as I said. So cool, cool fish, the two cool fish here. I can't remember what we talked about like 15 minutes ago. Hi, I'm Dory. Awesome. And with that, let's take a bite. Bye-bye. Bye-bye.