Join us in our latest episode where we delve into the dynamic world of agile software engineering with three renowned experts: Henrik Nyberg, the AI Whisperer; Nicholas Modig, a celebrated author; and Mattias Altyn, an accomplished agile CTO. They share their inspiring journeys, revealing how they transitioned from personal passions to spearheading innovations in tech and business. Peek into the successful Spotify model and discover how AI is changing our approach to teamwork and software development. Our guests weave their personal tales with practical insights, making complex topics accessible and engaging. Tune in for an enlightening mix of personal experiences, tech expertise, and innovative approaches to agile work and continuous improvement
Follow us on youtube: https://www.youtube.com/@aiawpodcast
Join us in our latest episode where we delve into the dynamic world of agile software engineering with three renowned experts: Henrik Nyberg, the AI Whisperer; Nicholas Modig, a celebrated author; and Mattias Altyn, an accomplished agile CTO. They share their inspiring journeys, revealing how they transitioned from personal passions to spearheading innovations in tech and business. Peek into the successful Spotify model and discover how AI is changing our approach to teamwork and software development. Our guests weave their personal tales with practical insights, making complex topics accessible and engaging. Tune in for an enlightening mix of personal experiences, tech expertise, and innovative approaches to agile work and continuous improvement
Follow us on youtube: https://www.youtube.com/@aiawpodcast
Okay, let's welcome you all here. We have three guests tonight. I don't think we had that before.
Henrik Göthberg:But we had I think we have sometimes when we did the what happened after summer, what happened after?
Anders Arpteg:when then we had right?
Henrik Göthberg:But not many times.
Anders Arpteg:Yeah, awesome, and also great to see some exporter people here, at least two of them here. So, without very welcome here, henrik Nyberg. I love the new title AI Whisperer.
Henrik Kniberg:Yeah.
Anders Arpteg:But I'd love to hear more about your background. But you're, I mean, so famous, I think, in all the extremely pedagogical and very fancy, and I'd love to hear more how you actually do this kind of visualizations explaining everything from agile and now, lately, the generative AI part as well Awesome stuff. We also have Nicholas Modig, right, best selling author and founder and CEO of Hubs, which we all work at the company. So I'd love to hear more about your background as well. And thirdly, we have Mattias Altyn great Also, agile coach, change agent, right. And CTPO.
Mattias Altin:is that correct CTPO, and technology and product officer CTPO I prefer CTPO, I think is better.
Henrik Kniberg:The more letters, the better. Right right.
Henrik Göthberg:And you were. You've been an engineering manager in different flavors, in both Spotify and background.
Mattias Altin:Actually, I lived in London for 18 years. I actually went to London 1999 to study for a year and then met my wife. She's Spanish, so we decided to stay in London for one year to figure it out and then turn it to 18. So I want to move back to Sweden to work first with Claude and then for Spotify.
Henrik Göthberg:Yeah, Claude, first right.
Anders Arpteg:Exactly briefly. I mean, we'd love to hear more about Hubs, about all the work that you do with AI and agile, etc. But first we should start with since you started, mattias, if you were to just give a very brief introduction to yourself who is really Mattias? What's your background?
Mattias Altin:Oh, yeah, I already started. Yes, I'm from North Sweden, lived in Gothenburg for 10 years and then ended up in the UK. I'm a software engineer since 97. In London I worked. You quickly end up in the finance industry there, that's the dominant industry. Yeah, and I worked for Hedge Funds National Management Firms and I was CTO for three different companies. So since 2008, roughly onwards.
Anders Arpteg:So really C3PO, c3po. Yeah, I love that C3TPO. I have my golden suit we're in office, awesome and at some point you met the other guys here, or how did that happen?
Henrik Göthberg:Yeah, how did you meet? That's a good question.
Mattias Altin:Well, Henrik and I live not far from each other at Faringsö.
Henrik Kniberg:So can I meet at the local burger joint.
Mattias Altin:Yeah, and actually our kids went to the same school then and I knew Henrik's wife, sofia, through the parent teacher. I love it, I love community and paddling Sofia's now with the kayak paddling. I actually also have a big interest in kayaking.
Henrik Kniberg:Big interest. He's like a hardcore pro yeah.
Mattias Altin:I used to be able to paddle, but I haven't kept it up.
Anders Arpteg:Awesome. And if we move over to Niklas, perhaps you can give a quick background about yourself as well.
Niklas Modig:All right, I'm from Stinungsund. Everyone have heard about the terrible accident in Stinungsund Either the bridge that fell apart more than 40 years ago or the road that collapsed a few months ago, so that's where I'm from. My background is that study. That's the Stockholm School of Economics, and that I pursued into researching and being a teacher there. I've always been a founder of Japan, so I lived in Japan as a kid and then I did my PhD studies in Japan and got opportunity to be.
Anders Arpteg:where came the interest from Japan?
Niklas Modig:I was into martial arts as everyone else, and I thought I was really, really fun and I had some naive idea that I'm going to be a karate kid. So I actually went to Japan as an exchange student when I was 16. So I think that's where it started.
Anders Arpteg:Was it karate or karate?
Niklas Modig:No, no, no. I had this vision that my Japanese host father, he, would like train me in the garden and be this mean guy. But, he played the trombone.
Henrik Göthberg:Wax on, wax off.
Niklas Modig:Now it was completely different. So yeah, I came to do some Aikido and some Kendo and stuff, but not so much.
Henrik Göthberg:Are you into it still?
Niklas Modig:No, actually not. I was into it until I was 20 or so, but when I moved to Stockholm I didn't continue because I felt it's a lot about your teachers and I had amazing teachers in Stilingsö.
Niklas Modig:And how did you fall into the whole lean Toyota production system angle that's talking about rabbit hole, but a short story is that when I let go of martial arts, I was dating a girl who was an aerobics instructor and the only way I could hang out with her was to take her class. So I started to take aerobics classes and then I got caught up because that was basically the same thing as martial arts doing hip-hop classes and step-up classes and so on. So once I just graduated from Stockholm School of Economics, it was one of my teachers at Niklas. I really liked your moves, you should become an aerobics instructor. So I declined my offer from becoming a management consultant and then I said I'm going to be aerobics instructor.
Niklas Modig:So with a master's degree in economics, I started to teach aerobics and then I thought the teaching processes are so inefficient so I have to apply lean here. So I created my own concept called Value Added Aerobics, where I basically applied lean on teaching methods. And that's what I told my old professor about, because he said, oh, have you been stopped? What are you doing? I do lean service within aerobics. And then he got hooked.
Henrik Kniberg:What was his?
Niklas Modig:reaction. No but he thought it was interesting and that's why he said why don't you come to the PhD program and write about lean service?
Henrik Göthberg:So my question how did you fall into the auto-production system? Is very accurate.
Niklas Modig:It's very, very accurate. Now it's true, I've never had a plan to take a PhD, or or I've never had a plan to apply lean. It was just that I had really, really inefficient teaching methods.
Henrik Göthberg:You straightened with inefficiency.
Niklas Modig:No, I was just starting to feel stupid and I saw people feeling stupid. How can I make them feel successful and get good workouts? And then I had to start applying how to make people understand.
Henrik Göthberg:that's a value-adding process and I guess you had had a little bit of lean exposure from the schools, from schools. So that's why you thought about it, because not everybody that is an aerobics instructor would think lean and go that.
Niklas Modig:No, no, I have my master's degree within operation strategy.
Henrik Göthberg:So that's it. You applied from school.
Anders Arpteg:And how did you meet the other two guys?
Niklas Modig:Well, henrik, since after I well, where should I start? Well, I pursued some research in Japan when I started to research about lean and services. Then I got the opportunity to continue at Toyota Japan and write about how have Toyota actually applied the Toyota production system in their demand chain, or sales and services.
Henrik Göthberg:So you were working inside Toyota. You could do your research in collaboration with Toyota, right.
Niklas Modig:Yeah, I got access through my university, so I was there two years and then I started to. Then lean was really hot. This was 15 years ago and the Toyota way was just published. Everyone was walking around with Lycose book and so on. So I gave a lot of speeches about lean and then I met some dude talking about Agen. I said what the heck?
Henrik Göthberg:is that? Yeah, but what?
Niklas Modig:is that encounter? What is?
Henrik Göthberg:that encounter. That's fun. What happened?
Niklas Modig:None of it. Talking about Henry we, I heard about him and then one day he said hey, I have a really cool woman here, marianne Popendick. She's a lean software guru. Why don't you come over for dinner?
Henrik Kniberg:And since then we have been good friends and the background there was that I stumbled over his book at my office. I didn't know who he was. I found I just and I was learning a lot of lean at the time I was interested in it. I found his book and I was like this is the best lean book ever. It was small, short and to the point, without a bunch of buzzwords. This is awesome. Oh, it's a Swedish guy. That's interesting. And I was collaborating with a lady named Mary Popendick, who's very big in the lean world, and then she was coming to Sweden and I'm like I want to get these people together and just geek out on lean stuff, and that's how we started.
Anders Arpteg:What's the title of the book? By the way, this is Lean. This is Lean. I add this to the list about topics to discuss and to know more about. You know what really write you someone?
Henrik Göthberg:read this book. Read a book and write the book.
Niklas Modig:This is a fascinating topic.
Henrik Göthberg:I'm always jealous when I meet people who have published books. I like it. But when was this, by the way, in sort of how we can go to you now, henrik Like? So the background here because now we're sort of getting into an intersect Is that when you are at Spotify or at Lego, or what are you doing at the time when you guys meet or you read?
Henrik Kniberg:the book. I think I was at Spotify, maybe at the time.
Henrik Göthberg:Yeah, it must be. So it's a while back.
Niklas Modig:Yeah, I would guess, maybe 12 years back, 10, 12 years back.
Anders Arpteg:It was a while ago, so who is really Henrik Neidberg? Can you give a quick backdrop?
Henrik Kniberg:I'm very curious about that myself. I hope to discover one day.
Henrik Göthberg:Actually, this is why you're in this pod. It will be a meditative stay.
Henrik Kniberg:I'm a very curious guy. I guess I actually, by pure coincidence also grew up in actually I grew up in Japan, so we have a common connection there. We don't know each other there, but when I was a seven month old baby, my parents all went to Japan because my dad worked there and then, yeah, so I spent 16 years as a kid growing up in Tokyo.
Anders Arpteg:You speak Japanese as well.
Henrik Kniberg:Not as good as he does, which is which is ironic, because I spent so much more time in Japan, but I went to an American school and spoke Swedish at home, and so yeah my Japanese is decent but kind of crappy by now.
Henrik Kniberg:But anyway, grew up there, came to Sweden it's a teenager decided that I wanted to be a musician and then decided that I don't want to be a musician because it's hard to make a living off being a musician. So what music instrument I play? Bass and piano and guitar and drums and can't decide what my we have so many.
Henrik Göthberg:We have so many podcast guests, so we need to. We need to have a jam, and jam we're talking about this.
Henrik Kniberg:There's instruments here.
Niklas Modig:Don't tempt me but anyway.
Henrik Göthberg:So I decided that maybe I have more instruments before. My keyboard was here before Now, so no but anyway.
Henrik Kniberg:So I decided that maybe I could play music just for fun and then have something else that'll pay the bills. So that ended up being programming, because I like programming and computers. So I started at, started at KTH and Stockholm and became a you know computer guy and ended up I got dragged out into various consulting engagements while studying and was coding out in different kind of projects, typically rather big projects like big telco projects and I was this one guy sitting there coding and I learned very much about how to fail with big projects. That was kind of. My big experience was that you need to fail to learn something. Yeah, I would spend time writing what I thought was really nice code, but then you know everything has to fit together and these big projects tend to fail and I was became a little bit of a cynic, like it.
Henrik Kniberg:But anyway, then later on I was in my own startup. A friend dragged me. I always get dragged into startups. I don't know why this keeps happening, but anyway I tried to start up and then I had had to build my own teams and then I didn't want to fail. So I knew exactly how to not build software successfully. Now I wanted to figure out. Is there a better way? So I started frantically looking around. Is there any information? Because by then there was this really cool thing called the internet, so you can look stuff up.
Niklas Modig:So here was this yeah 95 98.
Henrik Kniberg:No, that 98 was when we started the company. So I started looking around and I started finding something that later on came to be known as agile.
Henrik Göthberg:No, it's the manifesto. When was that written? 2001? This was before. Yeah, it's exactly yeah.
Henrik Kniberg:So there's this community of people who later on wrote the manifesto, but they created the first wiki because they actually invented the wiki. That's guy named Ward Cunningham invented the first wiki and then through that community how I don't know how I stumbled into it, but they were sharing knowledge like how you know working teams, right, Code that tests, code working pairs, all these really weird practices. That was very different from what I had seen in the field but that seemed to make sense. So I applied that with my teams and was like, oh it works, this kind of works better.
Henrik Göthberg:You know it's not a silver bullet. It works better.
Henrik Kniberg:It works better, I like that no guarantees, but when we fail, we would fail for the right reason and not because our process is stupid, right? So anyway, that's how I got kind of hooked into what later on came to be known as agile. And then gradually, through my mixing between consulting and startups, it started becoming like one of my tools, like, okay, I'm a guy who writes code, but I also learned a few things about how you can collaborate. And then more and more that became my image. And then I got involved in companies as kind of a coach and then started writing books and stuff around that. And then suddenly people were calling me something. They were calling an agile coach.
Henrik Kniberg:Agile coach, yeah they were accusing me of this. They gave me this title.
Henrik Göthberg:Yeah, did you want it or do you want?
Henrik Kniberg:it, it was, I don't know.
Henrik Göthberg:I remember how we started talking about agile coaches and now it's almost gone full circle. I don't want any more coaches, I want to read coders.
Henrik Kniberg:No, I was more like is this what we call what I do than fine, it doesn't matter. I was interested in helping companies improve and this really was helpful, so I felt good going to a company and leaving them in a state that was better than before.
Henrik Göthberg:And your years is fortified. Are you an agile coach there? Are you a line manager, or what different roles did you sort of work in?
Henrik Kniberg:No, early on in my career I developed a skill of dodging formal roles. Yes, I've been CTO three times and decided that I'm not a good line manager. I really prefer to be that guy on the side who throws in good ideas but not actually have stuff.
Henrik Göthberg:I need to learn from you. This is more me. No people person.
Henrik Kniberg:So that's kind of, yeah, spotify I was a consultant, but I was there for maybe six years or so and a long time, and most people forgot that I was a consultant. Yeah, I know, but I was really just a guy, an ideas guy. I was involved very early through random coincidence. I did a talk at KTH where one of the first Spotify teams was there and they got really inspired by what I was talking about scrum. So they started applying scrum and that's how I ended up coming in there as a kind of coach. So, yeah, helping them figure out as they grew from just a few teams to more people than how do we collaborate effectively?
Henrik Göthberg:So you guys did at the same time. Who was? Who was it spotless for? At the same time.
Anders Arpteg:Yeah, you had to. You left that 2015, right.
Henrik Kniberg:Yeah, something like that.
Anders Arpteg:And then I am at 17.
Mattias Altin:Same as me. Then I joined in spring of 17.
Anders Arpteg:Are you joined in 17?
Mattias Altin:I left at 17.
Henrik Göthberg:And where did you start?
Anders Arpteg:In spot 12.
Henrik Kniberg:Yeah, so you would have been there overlapping a little bit, for sure, yeah, and that was an interesting case because I was a company that was kind of, in my mind, born agile, so I didn't have to transform them from something else. They were born with an agile mindset, which was super fun for me to. You know, I could affect a lot more.
Henrik Göthberg:And we have another rabbit hole here the famous Spotify model.
Anders Arpteg:We have it in the list.
Henrik Göthberg:I remember vividly. I've been trying to explain this. We didn't know each other, but look here. Hendrik says this is a point in time.
Niklas Modig:It's evolving.
Henrik Göthberg:It's a snapshot, it's a snapshot and no one read the memo. No, no, no, but anyway, so yeah long story short.
Henrik Kniberg:I started getting involved in companies that like through my writings and stuff. I was lucky to get involved in some interest in companies like Spotify and Lego and did that for a while, but then at some point I kind of did a pivot into climate.
Henrik Kniberg:So I started getting panic about climate change. So I did a pivot and just dived into climate change and was involved with some startups there and made another fancy video about that. And then through another just whoop I happen to end up in Mojang as a Minecraft developer through some really crazy turn of events.
Henrik Göthberg:So you had a little talk in Mojang. How long was that?
Henrik Kniberg:I was there for about four years, four years.
Henrik Göthberg:More than a total, more than like a foot, yeah, so first, as a coach, and then as a developer and designer.
Henrik Kniberg:And then we started. We started up hops as part of that, and then, and then should we move that?
Anders Arpteg:perhaps?
Henrik Kniberg:Yeah, and then the last thing, of course, the whole AI thing, which I guess we'll be talking more about.
Henrik Göthberg:So, yeah, I don't know what the switch to guys. We need four hours.
Henrik Kniberg:I can tell already.
Anders Arpteg:Before we move into the topic of the origins of hops, can you just elaborate a bit? How did this switch to AI come about for you?
Henrik Kniberg:Well, chat GPT came along as a thing that way or yeah, all right, cool, I'm being moved around a little bit, now I'm in a better position. So, yeah, chat GPT came along and kind of started getting me excited. I've always been interested in AI. I did a bit of AI programming in Minecraft, but also from way back in time. I've always been kind of interested in this and, yeah, it was seemed like a really cool thing, but more like a toy than anything else. But then GPT-4 came along and when I started looking into that I was shocked to the core. Like I realized what I'm seeing now is something that I'll never again experience in my life. So I kind of, from one day to the next, made a pivot and said this is going to be the main thing for me now I need to figure this out.
Henrik Göthberg:I mean we need to, and this will have so many profound implications back to team and organization.
Anders Arpteg:I mean, you know, just thinking about teams that are agents, that are autonomous AIs next to people teams as an interesting topic in itself, yeah, but I had a topic as well about the generative AI video that you posted, so let's get back to that later, and I'd love to hear the background and how that started as well. But to the next topic how did Hubs get started? What's the origin? Why did you get to come together? What's the purpose of Hubs?
Henrik Göthberg:Who wants to start? I think, niklas, you start this one.
Niklas Modig:I can start. I mean the foundation of Hubs. I would say that maybe all of us have had a growing ID in parallel. My journey, that was integrated first with Hendrik's journey and then with Matthias. My journey started in 2011 when I was working with one of the sub companies that were producing radar systems and I had a lecture with one of their management team and they said, oh, we're about lean. And they said, well, we don't really want to implement lean, but hey, we're 1500 people and we're sitting in different offices and lean is a systems theory. So if this is really going to work, we need to involve everyone. So how do we do that? And then I said, hey, I know a camera guy. He could come here and he could record our workshop. So let's have a workshop with your middle man years. There were 70 people and then I brought in the camera guy, so he got some instruction.
Niklas Modig:I'm gonna have nine modules, so take the recording and record them, and after that, in the post editing, I wanted to take away all the discussions. So each of these nine modules, they were a little bit like first a story, then a question, then brainstorming and discussion, so very simple. So we had that workshop and then we cut up these nine modules and burn it into DVDs because we couldn't stream them. So we took all of the managers, got two DVDs each and then we said, okay, now you show. You have a timeline of four and a half months, so show one video each every second week and show the video and then discuss the question and you lead the discussion. So all the managers did that. So first module, second module and third module, and then basically all the middle managers educated 100% of the company at the same time. So everyone got the same story, same mindset.
Niklas Modig:And I remember after three modules the CEO called me and a guy from Gattenberg, anders, he's like what the heck have you done, moodig? People are right crazy. So he said that we have registered 57 different improvement initiatives for the first and second modules From the employees. That wasn't asked by the manager what has actually happened, they starting to break the silos. So he was very excited, but he said we weren't really ready for this.
Niklas Modig:So, and then they did a tremendous transformation, much, much faster than anyone that I've seen. And I was like, hey, this is interesting. And outcome there was like, okay, what did we do here? By coincidence, we made the managers to own the transformation. We made the managers into expert. You know, when I take Hendrik's new AI video and I show it to my wife, I get the credit because I don't know about that, because I came with a video. And that's the fun part, even if it's Hendrik, I came with a video and that's the phenomenon that I think is interesting that all the manager, they came with a video I'm going to show you and they became the expert who come with the message. So we developed 70 really great managers into experts and coaches and they felt that this is our transformation Owned internally, completely owned, and the engagement came from the managers and all the employees were also educated exactly at the same time.
Henrik Göthberg:So there are many very simple rules of thumb here that I think works together. That makes a huge compounding effect on this Huge yeah, involving everyone. But then it led to you sort of thinking OK, oh, this is apparently not everybody's doing it like that. We should think about this.
Niklas Modig:No. Then I started to experiment and then we developed and then when it was possible to stream, then I had third party LMS to spread videos and so on. So I was doing experiments for many years and after some time I realized that if you actually do this, this generates a lot of IDs and then an LMS is not enough because we want to take that idea, and that's a learning management system.
Niklas Modig:Exactly, Learning management system, so a simple e-learning system.
Niklas Modig:But if you do that in a company and everyone talks about the same question for instance, how do we decrease lead time and you have 1,500 IDs how to decrease lead time Then you want to take that and condense that into a strategy that you actually go from IDs to action Dispensability of data, Exactly. So we needed something more advanced, and that's what I felt like hey, we need to develop something that is different from LMS learning management system. So that's why we started to develop a transformation management system, TMS and that was the starting point of HOPPS. And that's when I was thinking because I loved thinking about storytelling and metaphors and seeing how can we impact people and I was thinking, hmm, who is my big ILO there, Henrik? So I called him up and said hey, I have this idea about founding a company where we are developing a software where we can involve everyone at the same time, but we're not only involving them in terms of training, we also generate improvements, IDs and make sure that they can implement it in a smart way.
Anders Arpteg:And what year was this?
Niklas Modig:We started it almost five years ago and then we found it as operationalexcellencecom, so it was like hotelscom, operationalexcellencecom, but then we understood that that's not a good name. So then we ended up with HOPPS, which is human upscaling and human upscaling, so involving everyone, and build the people. So that's the simple, both vision and idea behind the company.
Henrik Kniberg:It's a fun, fun story there. When he called me, I was in my summer cottage and he was in his summer cottage and I get really inspired by it. He was like we should talk about this. Yeah, we should talk about it. Where are you? Oh wait, You're just, you're not so far away. And so he got on his boat.
Anders Arpteg:And just came to my house. I love it.
Henrik Kniberg:So he took his boat to my place and we're just chatting. It kind of just felt an instant, all good ideas needs to happen through a boat ride.
Henrik Göthberg:Yes, that's another rule of thumb, right.
Niklas Modig:You can take over, so where did you start?
Henrik Kniberg:Oh no, it was basically that, but I think when he called and it matched something that I'd been feeling for quite a while, what was that match? The match was that I had done a bunch of like Nicholas, done a bunch of talks, done a bunch of talks and keynotes and written articles, taught courses, stuff like that and coached, but then it's me spending a lot of time on site with a limited group of people, one company, you use a bandwidth problem.
Henrik Göthberg:Bandwidth problem yeah.
Henrik Kniberg:And my impact is limited. Sometimes I can make a big impact, sometimes I spend a lot of time and make no impact, but then I would put out a video, for example, the Spotify video as an example.
Henrik Kniberg:And then the video goes crazy, viral, and then we have hundreds of companies making big changes without completely out of my control and I'm like, hmm, something in the middle here, yeah, like maybe I don't have to be standing on the forefront, maybe we can scale this and not just release a video in the wild there's hope for the best, because that doesn't always give the best results but also not be the bottleneck guy. And I think what Nicholas was talking about just resonated with like yeah, we can do get the best of both worlds. Kind of Interesting.
Henrik Göthberg:So there is a way of working, methodology, organization, ways to transform, but there is also technology part to this in order to figure this out, to becomes productized and scaled. And I guess that's the angle this guy come in, matthias, or how did you come up? Because you joined the guys a little bit later.
Mattias Altin:Yeah, Six months now, right, yeah, I got a call from an ex-colleague from Spotify, jonas Saxisson. They say, hey, henrik Kneeberg and Nicholas Moody wants to have a chat with you and they're doing a startup. And I was like, oh, startup now Not really interesting. But Henrik is a nice guy and heard good things about Nicholas, so of course it ended up being a breakfast in Henrik's living room.
Henrik Göthberg:No, both trips at that time, no I came on a motorbike because I was going to drive into Stockholm. That works.
Mattias Altin:That works.
Niklas Modig:We did it early.
Mattias Altin:I have a meeting in Stockholm, so I might as well just stop by and have a chat with some nice people. And then that kind of led to a number of meetings and I was really not convinced at all what.
Mattias Altin:I understood in the beginning was like ah, I don't think so, but the more we talked and we did that, we had the. We met few days later in the local golf club restaurant. I don't play golf, but Nicholas taught me how to swing a club in a simulator and we were talking. That's the back problem. Come on, don't blame your fucking bed Come on in Japan.
Henrik Kniberg:That's the story you tell your wife.
Mattias Altin:I get that. It's the bad golf swing though, and I've never played golf in my life, but I was the first to know the time I had a business meeting with golf, and then we kept talking. I then kind of thought it's not for me, but I kept thinking about this. I entered some other processes for jobs more traditionally jobs maybe but I never felt excitement about that. But this thing kept going over in my head and we kept talking. Nicholas and I probably spent 20, 30, 40 hours total and eventually outside. Now that's what I want to do. It's, I think it's a unique proposition. We have to really kind of create what Henrik described, a kind of self-improving.
Anders Arpteg:Can you describe that? What's the typical customer of Hubs?
Mattias Altin:I think Nicholas and Henrik are probably best for that.
Niklas Modig:Typical customer I would say it's a large Swedish industrial company or international company that have probably been working with some kind of improvement initiative, with operational excellence or lean on Agile, and they say that, hey, we really need to scale this and then we can help them. So everything from smaller companies with a couple of hundreds and employees but how large is customers? They have more than 20,000 people. So global companies that really want to scale training and scale transformations, and with a deep, deep thought that we want to make our employees own and do this, because I think that is. There are different ways of driving transformations, but I think building in the competence into the leaders and making them take full responsibility of everything, I think that's the leaders or the employees, both. First the leaders, so they can develop alignment and then they involve them, please, because you need a crystal clear alignment, otherwise it won't work.
Henrik Göthberg:And I think this speaks to the core when I talk about in their backs, of becoming data and AI already. Don't talk about the tech, don't talk about the AI, talk about owning the shit, talk about who owns the data, who owns the tech, who owns the algorithm and I'm super scared of this. I start calling it hero consulting. I've been doing that myself and it's not good when, literally, you have someone coming in and you're fixing stuff and it's literally a black box for the organization and it will never be sustainable. And ultimately, it comes down to ownership owning your own chain, owning your own transformation, owning your own capability.
Henrik Göthberg:So it really resonates what you're saying. It really resonates and I think a lot of people they are using consultants in the wrong way, even.
Anders Arpteg:Yeah, I mean awesome, and we promised to get back to your book as well. What is lean right? But was that the right title?
Niklas Modig:The first title was actually what is lean. Then we changed it into this is lean.
Anders Arpteg:This is lean Question and answer Good. Ok, please describe a bit what's the I guess core messages of this is lean.
Niklas Modig:But since I was doing my research in Japan, I got opportunity to go into Toyota Japan and see what is their version of lean meaning the Toyota production system, and lean was really, really hot at that point. So everyone was talking about lean tools and lean methods and everyone was going to do lean and become lean, and it was both a verb and adjective, so it became really, really messy.
Henrik Kniberg:Everything was lean and buzzword party. What is lean?
Niklas Modig:So, but when we came to Toyota, there were no buzzwords. They say, hey, it's about flow, it's about customer as a need, and the only thing we do is to fulfill that need in the best way, as fast as possible, and we'll wait until we have the cash. That's the only principle that we're focusing on.
Henrik Göthberg:Flow.
Niklas Modig:Flow Nagarizu-Guri in Japanese, and that is basically yeah, just let the water flow.
Anders Arpteg:So that's a lot of training as well, yeah.
Niklas Modig:I mean, that's one principle of the two, the flow thinking, and basically I love this metaphor that they taught me, like if you put up the film camera on the customer's shoulders, how can you make sure that the customer journey is great? Or if they have a car and the car needs to be repaired, from the point when the car needs to be repaired, put up the film camera on the car and see what is happening with the car until it's repaired. So it's basically the perspective of the flow unit and that is applicable in every single industry, irrespective if it's manufacturing or materials, information, customers. It is about the flow. So that's the first principle and then it's the Jidoka principle, which is, basically, if you have the flow, how can you make sure that all the employees can see the flow and know what is the normal flow and if it deviates, everyone react immediately and correct it?
Anders Arpteg:What was the name of the?
Niklas Modig:principle Jidoka, jidoka, jidoka.
Henrik Göthberg:So there's two Japanese terms. That is the core. Say them in Japanese.
Niklas Modig:The first one is, just in time, they call it, but in Japanese, nagaresu-guri, so it's basically to create flow. And then it's Jidoka, which is it's three different characters self-working, isation, meaning how can we create the culture where everyone knows what to do? All the time, that is done through visualization, and that's why everyone knows what is normal and then can see what is abnormal. So it's a very conscious learning system.
Henrik Göthberg:But it is not interesting, it's not easily translated, it's like semantics, and when you have Japanese characters and how that works, and then you want to try that and put that in English words, I can see how much more powerful those characters, that sort of built this whole. You know you have a whole spiel to explain. Is that true? I mean like, so is it more powerful in Japanese, these two works or these two characters?
Niklas Modig:I think it's also some, you know, if you say oh, this is yakiniko and it sounds better, but it's basically fried meat. It's not so sexy, but it sounds good. Yakiniko, you know, but it's grilled meat.
Henrik Göthberg:Take away the curtain. Boys, Take away the curtain.
Henrik Kniberg:I like it, but this actually is one of the kind of Nicholas things which also I subscribed to, which is let's reduce the buzzwords and just talk about the thing itself, I like it too.
Anders Arpteg:OK, so back to the question of what is lean? Then if you were to try to just condense it to the core, that's the two.
Niklas Modig:How can a company create flow? Put up the film camera on what they are producing, and how can we create a system so everyone sees what's happening all the time, so they can react on deviations immediately? If it's a hospital, then you put up the film camera on the patient, then you focus on the patient's flow and then you want all the personnel to see. Can we deliver a normal flow from the patient's perspective? Not writing a referral, not sending them to other. It's the patient's journey.
Anders Arpteg:If I phrase it like this, what's the opposite of lean?
Niklas Modig:Silent thinking. Put up the film camera on the machine or on the doctor, make sure that he or she is busy so we optimize his or her flow instead. Then we have the film camera on the doctor. We make sure that he or she always have something to do. That's opposite of lean.
Henrik Kniberg:And also a very common model.
Niklas Modig:Yeah the most common model.
Anders Arpteg:That, I think, really nailed it really well. I think the film camera metaphor is excellent.
Henrik Göthberg:Who are you putting the film camera on when you're trying to optimize or try to keep people busy?
Niklas Modig:Yeah, and that's for me. That was the connection with the HRL as well. Oh hey, we have a need for some kind of new functionality. Where do we put up the camera until the functional is developed and used and we have engagement there? That's the flow, so it's the same thing.
Henrik Kniberg:So instead of a physical gadget, it could be a stick in the wall representing a feature, but the camera is on that feature. We're optimizing for getting that, moving from idea to production.
Anders Arpteg:Awesome. I'm eager to move to the Spotify model already. Should we go there perhaps, or is it too soon?
Henrik Göthberg:Let's go there.
Henrik Göthberg:We're exploring different topics. Now, do we actually have a theme for this? We've been trying to do in 2024, this season and a little bit last season like, oh, we need a good clickbait theme for a podcast. Now we have a very distinguished guest, so you're the clickbaits. That's okay. But if we flip it, what is the theme here? We are talking about hubs. We are talking about you coming from different angles here Agile, lean engineering and ultimately now we're moving into an AI world. We are pivoting, but what's the underlying interesting, core topic here? I think it's about transformation. Maybe how do you get flow and adoption in transformation and, ultimately, how do you get flow and acceleration in adopting data and AI? Maybe that could be one angle, or do we have another one, I think?
Anders Arpteg:we formally wrote the theme as the role of AI in the future of software development or something like that, but I still think we should. Before we go there, just take this Spotify model.
Henrik Göthberg:I think that we are building up to a theme by looking at different perspectives into this story.
Anders Arpteg:We're looking at into a Lean story perspective and now let's look about I mean, this Spotify model became really popular, and not the least because of all your work, henrik as well, and the videos. That, I think, is one of the most popular ones when it comes to Agile work as well. Perhaps you can elaborate a bit more what was the idea when you started to make that video? And just go continue to describe. What is this Spotify model?
Henrik Kniberg:Yeah. So I was there in the early days and I noticed this amazing culture, which was different from any place that I've ever seen, and I was fascinated by it. And also many of the people around me just felt this is different. And as time passed we grew very quickly and it often became my role informally to help onboard people who came in, especially when new coaches came in, because they're supposed to be culture, help spread culture.
Anders Arpteg:Perhaps you should explain that they don't have Scrum Masters in normal census body fight. They had Agile coaches.
Henrik Kniberg:At the time, when it just started, they didn't have anything like that. But then when we decided to start using Scrum very early and they had Scrum Masters and then quite quickly and I started coming in a little bit we decided to let's not narrow down too much on Scrum, because Scrum is one flavor of Agile. So we basically just renamed the role. So the early Agile coaches at Spotify were essentially Scrum Masters but we called them Agile coaches because we don't want to make everybody feel like they have to use Scrum specifically. That was the only real reason.
Henrik Kniberg:And Master doesn't sound good either Scrum Master might make you ask who then is the Scrum slave?
Henrik Göthberg:And then you know it gets icky right.
Henrik Kniberg:I'm not very fond of the term at all, but anyway. So I started spending a lot of my time explaining to people, trying to verbalize what is it that makes what is our culture. And then, at the time, also the CTO. At the time, oscar was doing a good job in Oskar Stol. Oskar Stol, yes, he was preparing for he was seeing this growth happening very fast growth and needed to create some kind of framework for us to grow into. When you're just two or three teams, you don't need much structure, but when we're going to be maybe 15, 20, maybe we need to figure out which teams need to talk to which teams. So we need some kind of a containment structure. And then it became just called a tribe. But essentially he was preparing, creating a structure that wasn't needed at the time but was going to be needed. So and I would give most of the credit to him, I was involved as a sounding board, but it was essentially mostly his work.
Henrik Göthberg:So the Spotify model and Oskar Stol is closely linked At least in my head.
Henrik Kniberg:But, the Spotify model is more than that, but specifically the squads, tribes, chapters, guilds, terminology yeah, mostly, but there was more to the culture, so I would say there's no one person. There was a number of people that together made this cake, but I became often the guy who visualized the cake, because culture is often invisible. I like to think of culture as stuff people do without knowing why they do it, and I became the guy who tried to make it visible. And then at some point I think it was Oskar actually who suggested that you know, henry, instead of spending a lot of your time running around doing the same presentation, why not just make a video out of it? So I was like yeah good idea.
Henrik Kniberg:So I made a video about it, and mainly for internal use, but we also figured we could put this out publicly, because Spotify has learned a lot from other companies. We stole a lot of ideas, so it wasn't really original to Spotify, but so it was like maybe we can give back a little bit by sharing what we've been doing. And so we put it out there and that was pretty much it. There was no plan. We didn't intend to make a model or anything, but it became incredibly viral and then actually became a good recruitment tool initially.
Niklas Modig:But it was deeply there was no plan.
Henrik Göthberg:It was not planning to make a model and the whole model also people inspired by oh, this is the way I would love to work, you know so that that's how that became a recruitment tool. I guess.
Henrik Kniberg:Yeah, I guess it's two things. One, it became a recruitment tool for Spotify, but, but most importantly, it inspired many, many other companies to change. Yeah, and my theory about that is I think a lot of companies wanted to change. They're in a position where they're not happy with the current situation, and I think this one case study became the concrete example of like it's possible to work in a different way. Yeah, so, whether or not they do the same as we did, they got inspired by hey there are other ways of working.
Henrik Kniberg:We don't have to follow the standard approach.
Henrik Göthberg:But this also highlights how important but also hard it is to communicate and visualize and put the finger on something, because sometimes a lot of we have an idea and we could kind of do it, but it's tacit knowledge. Yeah it's not real knowledge explicit, it's not done in a way that is sort of approachable, and I think that's that's a huge thing. You succeeded with that simple video.
Henrik Kniberg:It's a bit of a blessing and occurs, I think you would like to compare with with the lean that when lean became a thing, then everyone started copying Toyota. But the important thing wasn't the practices, it's the principles behind. Yeah, and the same thing there. People looked at the Spotify videos and I had a lot of examples of specific things we did. What I think was interesting was the underlying principles and values, but it's always easier to copy practices, yes, so a lot of people just be like, hey, we'll just do exactly what they're doing and without pirating the practice, without subscribing to the principle, yeah, which is similar to a lot of companies that kind of do lean cargo culting.
Henrik Kniberg:They apply the practices but the culture itself isn't really there. They're still putting the camera on the doctor, that kind of thing.
Anders Arpteg:What is the name of the video? You had made a lot of them, so which one? That?
Henrik Kniberg:was called Spotify engineering culture, and a really important thing to emphasize is that was a snapshot in time. It's not intended to represent Spotify today or anything else. It was really just here's how we worked at the time.
Anders Arpteg:Was it how it worked, or was it how you, we, wanted it to work?
Henrik Kniberg:That's actually an interesting question. When I made the video, I was paranoid about that, because I really wanted to make sure that this is something people can feel is correct. Yeah, but Spotify was also very ambitious. They're always wanting to improve, so I finally ended up making a video that was a bit of both. It's a bit of how we work and a bit of how we want to work. Yes, and when I tested the content, when I talked to teams to check, is this accurate, what I typically got from any given person or leader, they would say hey, henrik, 70% here is spot on, this is exactly our culture and how we work. 20% is like maybe, sort of, and 10% is bullshit. No way, nowhere, nowhere near. But the funny thing is, different teams had different 10%.
Henrik Kniberg:So on average my conclusion was, yeah, this is a fair, a good representation of how we work and kind of how we want to work, but not 100%. But of course, as time passed, you know, spotify kept evolving and this is the really important part that the magic of Spotify, in my experience, was not that exact way of working. It was this culture of continuous improvement that was really baked in.
Henrik Göthberg:So they continued experimenting and I'm going to say Matthias was there, you know more than you could probably talk more about the.
Anders Arpteg:What do you think? Was it how it worked or how we wanted it to work?
Mattias Altin:No, I actually going back to the first time I met Henrik, I was really nervous and thought this might get awkward because I didn't know him. That how great it is. But I, three months into my tenure at Spotify, was tribly for a tribe there and there have been a lot of commotion around the tribe for two years. I wrote a white paper explaining why I think my tribe, or our tribe, should try to try something else, not try the Spotify model, and it was based on data and induce and so on. So which we actually went back to having more folks on leaders that were present, so an engineer leader that actually have organizational health, technical health, delivery and execution as accountabilities being present with the team. And we also went from product owner to product manager in the team, in the squad.
Anders Arpteg:So the squad had a product manager instead of a product manager A product manager, yeah, instead of product owner.
Mattias Altin:And what does?
Henrik Göthberg:that mean that difference in semantics. I will pass that to Henrik.
Henrik Kniberg:I'll pass it right back.
Mattias Altin:It's a bit of a change you made More, maybe being a master of ceremony and standing every stand up trying to feed the team with tickets to dues and be more of a strategic finger, looking ahead and taking like a product strategy.
Anders Arpteg:Some people put like product owner and product manager in like a hierarchy and say that the product manager is about Pass.
Mattias Altin:I for me is more of a let's try to have the EM and your manager being more present with the team owning organizational health, which was to hire, recruit and recruit, mentor, coach, making sure that the team actually is competent. And then the second one, technical health. Actually, you should know a thing or two about what you do and be able to. Then you could be a good mentor but also be able to figure out what is what are we building? When should we take on technical debt? When should we pay it off? And then, with those two levers, you can actually own the lever and execution as well, because you know when you can go faster, why can't we go faster? We're unstaffed or we're under skilled. So I wrote that white paper and was only intended to do as an experiment locally, and that turned into quite a big thing because my manager at the time said, hey, you should spread that to the whole spot.
Henrik Kniberg:I don't know I'm still on probation.
Mattias Altin:And it was like a lot of people loved it and thought that was great. We should try something different, and other people got really upset. Some people got upset but in the end of being, we got exempt to try this in September or December 2017.
Anders Arpteg:And what was the big change really?
Mattias Altin:Well, we took away the matrix system. So having chapter leads that say your manager, your expertise is back end, you have to be at the 50 in the X and they're all spread out over different teams. So when I went to my link I was the director saying, hey, how's it going in that squad, how's the delivery going? And quite often I go, I don't know, I am not close to what they're doing.
Anders Arpteg:We should just be for people that's not familiar to the Spotify model. Describe a bit what the tribe, chapter squad etc. Is.
Mattias Altin:Henry is not a witty about it.
Henrik Kniberg:All right, yeah, but yeah. So the Spotify model is, first of all, a lot more than just that, but I can make the main thing people mostly talk about. The picture I drew is basically a matrix structure. It's just a fancy term for a matrix, but squad is basically the rough equivalent of a scrum team or an agile. It's a delivery team. Their job is to deliver stuff, but it's pretty much maps to a scrum team. Five or six people, cross functional, have an end to end responsibility. That's one dimension. That's the main dimension. Your main home is the squad, but then there's different skills in there, such as backend engineer, maybe tester and stuff. So that's the chapter. So I may be in squad A working on, let's say, playlists or something, but I happen to be a backend engineer, so I'm also in the backend chapter and I'm talking to other backend engineers and other squads. So that's the matrix. But the difference is that squad is my primary home, so that's what squad and chapter is. Is that also called the reverse matrix in?
Anders Arpteg:some sense Kind of yeah.
Henrik Kniberg:Reverse matrix, because the power is in the squad mainly. The chapter dimension is a support structure, but it's also at least was also the line management structure. So my line manager would be another backend engineer in my squad or in some other squad, but that way I can switch squad without changing manager. That was kind of the idea. And then, for scaling reasons, then we had this higher level container called the tribe, based on the Dunbar's number, like if you're more than a certain number of people, you start losing connections with each other. So we don't want to have more than, let's say, 100 people or so in a tribe. So tribe would be a handful of squads and then there'd be a tribe leader and you were a tribe leader. But then some interests, some areas such as, like these chapters, like backend engineer, you might want to talk to backend engineers in other tribes as well.
Henrik Kniberg:So we introduced this more vague concept called guild, which is kind of like a community of interest. Anyone interested in agile coaching would be an agile coaching guild. Anyone interested in AI or data engineering would be in that guild and that's really just a community of interest. That was pretty much. That's just maybe 10% of the model, but it's the one that most people talk about, and I would say, in the less successful copies of Spotify model, they just copied that. So they say you know, you're an hour chapter lead. Oh, what do I do? Same thing as before, but you're a chapter lead and your team scrumpties is now called a squad, because now we're really cool, no damage done. Really, things don't get worse than before, but they also don't get a whole lot better.
Mattias Altin:I think the observation was that we didn't get high performing teams, because if you move people around like that you with Bruce Tuckman and Susan Willins Ferris, so that was like an obsession.
Mattias Altin:No, high performing teams and we also have data on chapter leads are leaving rapidly.
Mattias Altin:No one wanted internally step up in the chapter lead because they felt that was a dead end for their career. So the data we had then but was in the whole of Stockholm at the time I think it was we hired exclusively outside was only two individual contributors that went over to be a leader, but at the same time I think five, four or five had to come back to the IC. So we had this like all leaders are coming off of outside, and the thing that made me think about this was like what's going on? Because I'm walking around all these super talented people, I'm average. Why did it take two years to hire someone full time for this job? That is, I couldn't figure out, and I think it was like in four years, one or two people had actually gone from chapter lead to tribe lead and that was the sign for me that like we have not got a system where only my limited view where we actually grow in the next generation leaders. That issue, in my words, you be a continuous flow.
Anders Arpteg:That I remember also that the chapter leads. You know that we're managing people in the squads. You know that they really didn't have an insight you know, how the sports were doing In cars. How can you?
Mattias Altin:If you have 15 directs, you have one to one's 15 hours and you're going to hire and do all that and then, at the same time, you're going to understand what's going on in 15 different but all in your narrow. I do mobile, you do back and also have time to code.
Henrik Kniberg:Yeah, yeah.
Mattias Altin:So it's in super human tasks.
Anders Arpteg:So it switched to engineering managers instead.
Henrik Göthberg:Yeah, but we had Andres Newman here. Andrias Newman, andres Newman, head of search, engineering, manager of search. This is now in the 2020s, but clearly a product manager engineering manager style for one domain. So now we're going. So I've been going nerdy down into the domain driven design approaches and agent even agent based approaches, with very, very clear ideas. So this is there, like stuff we're talking about now you know, and then you can take it into data mission technology and all that but hardcore domain and thinking carefully about you know how teams serve other teams, so we should have a team talk.
Anders Arpteg:Let's continue speaking about the.
Henrik Göthberg:Spotify model. First Rabbit told. I'm the rabbit told guy, by the way.
Henrik Kniberg:Yes, I love the talk. Oh, actually I have one thing that would add to, since you mentioned beginning. You were a bit nervous to meet me and I suspect it was because you're like, but we didn't exactly follow the model. Maybe Henry could be a designer, or was that I?
Mattias Altin:wrote that paper right and it was shared. Oh, I was asked to share it against my will. And then, as I said, some people are really respected and looked up to got upset about it and I thought like, have I done something bad here? And then I think, like a week later, we met in the local burger with the kids I think we're going skiing, right, Henry's wife teach skiing in the local ski school. And so I was like, oh, should I tell him? Maybe it's really fendant? And I said, hey, I'm really sorry.
Niklas Modig:I changed it. Yeah, no, no, I didn't change it. I said are we going?
Mattias Altin:to run an experiment and we said we're going to measure like NPS. We had the lowest EMP score in that part of Spotify and then we went.
Anders Arpteg:We need to, I think, explain NPS. So net promoter score yes employee net promoter score.
Mattias Altin:How likely we had a really low score. Oh, he said that I was just cheering.
Henrik Kniberg:I'm like that is the Spotify model. That's what it is.
Mattias Altin:The relief when you were like oh, that's cool, that's perfect, that's the way it should be.
Henrik Kniberg:And in fact, if you had told me we're still working the exact same way as you wrote in your video XT years ago, then I'd be disappointed and worried, and that was only my situation at that time.
Mattias Altin:But what did happen? Actually, we did do this change and it was like a change in the role descriptions and we never got to really formalize. A big nervousness was around. Am I technical enough to be an engineer manager? That was a constant question and it also was a lot of maybe unfair agony against gathered coaches.
Anders Arpteg:But what is the real difference you mean between a chapter leader and an engineer manager. It really depends on the individual.
Mattias Altin:So I think most of the chapters were excellent and it's more like why did those people who had been an engineer, who wanted to become a leader and took the chapter role, want to leave and were really unhappy? It was basically what they said. I'm just a glorified people manager now I worked 10, 15 years to become a really strong engineer, and I haven't touched code or looked at any delivery for two years now.
Anders Arpteg:I don't like this.
Mattias Altin:So, as I said, I observed a very, very small part 12 of the I think it was 12 tribes at the time, so I can't answer the rest, and I heard most people, many of people, say that it worked perfectly. So I believe that's it, but do you guys have the view.
Henrik Göthberg:the sports find model was a snapshot in time and that was made when 15? 2014. 2014, I think 14, 15.
Henrik Kniberg:I think it was 2014. You?
Henrik Göthberg:know, and then when you left Spotify, if you would sort of summarize, because now we have grown, now we have maybe even more engineers in America than in Stockholm, right? So how big. You know what was the trend in terms of organization in Spotify at the point when you left?
Mattias Altin:So what happened was that we were the first tribe, I think, to do this formally like, and we got the exemption like okay, you can call it a new manager.
Mattias Altin:No 17, right, late 17,. Yeah, it was in September. I wrote this paper. We had an offsite in Stockholm with the tribe, but then we started doing this in 2019. And I was in the conference an agile and leadership conference in New York in September 18. And then it was a big uproar because then, all of a sudden, I think most of the tribes had done this to change and there was a lot of really scared people like Shepleys was doing the thing to protect you enough, and it was agile coaches. I was like what am I going to do now? So I think it was a toothbrush kind of swinging away from that.
Henrik Göthberg:Yeah, the transition to one thinking to another, I think it was like maybe an overreaction.
Mattias Altin:I said we did it very gradually, so how am I going to make the change? Well, hopefully, as an engineer, you will not notice the change. We will do this very gracefully. But what you are going to notice, actually, your manager, when you sit in your 101, she will actually be able to give you valid feedback because she was there at the time when you discussed an architectural decision. That will be maybe more than you might be used to, because your manager sadly is not present when he has 15 employees.
Anders Arpteg:So you've been really good at grabbing the problems, I think, with this butterfly model.
Mattias Altin:It also cleared Tyson Singer, who was my manager at the time. He wrote another document to help me with this to explain that we need to have flexibility. My tribe at that time was a platform tribe. We had a fixed goal. We had long-term. We weren't supposed to supersize the tribe because we were going into IPO and we kept our costs stable. Other tribes that work more on the forefront of developing consumer products they might need this model because it would be flexible. So it worked for RA.
Henrik Göthberg:we think I wouldn't say that that would automatically mean this right for everyone, but I think it's Henrik, I think I remember when I watched the video you were quite clear and this is lost in the memo that you need to carefully understand the context and you need to move the context. These are principles, these are ideas. The principle and ideas are more important than the practices.
Henrik Kniberg:They're principles and examples of how they were applied at this time at this company. Yeah, exactly. But I can add one thing about being dissatisfied. There's a general pattern I noticed which is interesting At Spotify. When we released that video, people outside were like, oh, this is a new religion. Yeah, exactly, this is Nirvana Exactly.
Henrik Kniberg:But internally at Spotify, people were always just complaining because not about the model, they didn't notice that. Most people didn't even notice the Rawa, they're just focusing on delivering stuff, but always dissatisfied with the status quo, and that's what made them succeed. And I saw the same thing at Lego. Yeah, I worked at Lego, a super successful company, always dissatisfied, and I just kind of get the same vibes about, maybe, toyota that I think the irony is that super successful companies maybe need the cultural element of always being dissatisfied with the current status quo and always wanting to improve. So everyone else was like, oh, those guys are so awesome. And the company themselves were like, no, everything is broken, we have to improve, which is kind of an interesting but that's a mindset in itself, right.
Henrik Kniberg:Isn't it that?
Henrik Göthberg:we are striving for excellence. And excellence in reality should be unattainable.
Henrik Kniberg:You should always be which also means you need like I had to coach people at Spotify to stop and celebrate a little bit. Catch yourself on the back, because you're always seeing the next thing to improve. But you've also made a dozen improvements and the companies doing really well and we're revolutionizing the music industry celebrate a little bit and then let's fix the next problem. Awesome.
Anders Arpteg:And I would recommend anyone that's interested in learning more about having the proper engineer and counselor to watch this video. Right, it's still applicable today, wouldn't you say?
Henrik Göthberg:I think it's definitely applicable as long as you see it as a general example and not a specific you know, I think that's the point Think about the words, think about the messages, think about the core values here, how you need to apply them technically, sort of thing.
Henrik Kniberg:I would say it's OK to copy paste another model, whether it's lean, whether it's Spotify, whatever it is copy paste, but also adapt Copy paste. And that's what Spotify did. We copy pasted it all over the place, but then we adapted. I think that's the main thing.
Anders Arpteg:Awesome. So, moving to some AI topic as well, I think, of course, HL is super important, but if we were to move into the world of AI that we have today and the AI first future that Google is calling it, how do you see? I think you wrote an article called something like the our developers needed in the age of AI, or something.
Henrik Kniberg:Yeah, that was it.
Anders Arpteg:What would you say the role of AI will be in the future software engineering field?
Henrik Kniberg:So I'm going to quote Kent Beck. Kent Beck was one of the early agile pioneers. He made a. He captured it perfectly in a tweet where he said something like this I resisted for a long time looking at GPT, but when I did, I instantly realized that 90% of my skills are now worth zero, but the other 10% are worth a thousand times more. So I need to be calibrated and I think that kind of captures it. I think software engineers are still going to be needed, at least for some near term but that the role is changing radically, and it's, I think, the same for other industries as well.
Anders Arpteg:But since you asked about software, I would say that I think that's a really good way to phrase it and if you take any kind of role, if it's customer support or whatever, you can get rid of, like perhaps 90% of the normal tax you have, but as a human, the 10% that's left that humans actually develop in AI becomes 10 times more valuable. So I think that is so. Many people are calling the future of AI as being more productive when using AI to get with humans, and I think that's a good example of saying you can focus on the 10% of the past and that, together with AI, will be 10, or a thousand or yeah, and also focus more on the things you didn't have time to do that you really wish you had time to do.
Henrik Kniberg:Now you have time to do them. That's nice Another way of putting it that I think someone said like what happens when we get more and more powerful tools.
Henrik Göthberg:If it's AI or anything else, we are allowed as a human species, as in our roles, to work on bigger problems. So one way of looking at your role or what you're doing is actually you need to move from the need to do things like problem, because that's what you give to the AI, but you need to work on the real problem, and now this is the 5-10% to figure out what is the real problem, what is the real question to ask? The AI is not smarter than the question you asked them or the task you give them, and this now becomes a high value commodity of humans.
Henrik Kniberg:I think a good comparison is in the past, in the not too distant past, programmers would spend a large part of their times on memory management, just figuring out where do I put the data and where do I put this piece of memory and what part of the memory you know, just that. And that was a big thing. And then garbage collection came along and coding, and now that's just automatic. That part of the work just completely disappeared. And nobody misses it, right? Because now we get to focus on what problem do I want to solve, rather than where does this byte of memory get put. So I think this is kind of a continuation of that.
Anders Arpteg:So is it like a level of abstraction that basically keeps increasing? In some sense? I think so. So you basically are telling I mean some people are even saying I mean I use AI all the time, or in chat, TBT and Gemini or programming purposes, and as soon as I hit some kind of problem, I don't do a Google search for it or I don't go to Stack Airflow anymore, I simply ask Gemini for the proper answer and it usually fix it, which is super powerful, of course.
Henrik Göthberg:But can I take the question? I can lead us into the question because I want to take the next topic, because now we've had a conversation that is essentially. The core theme of the conversation has been organization. It has lean and what that is all about is agile. It is how you changed. So giving the understanding that we need to focus on the 10% rather than the 100% sort of thing. That's where the human impact is, in, the 10% of what we're doing today. But that's 10% is 1000 times more important.
Henrik Göthberg:Where does this leave us on how to organize in these topics? Where does this leave us in how to think about teams and how to think about that? And we can, we can. Does it change anything? I mean, like we have AI, we have data scientists in the AI manager, we have different disciplines. Is there any change In Spotify? Little change, if I. If I compare to Scania that I've been working quite a bit with, they're fairly mature and then other companies where there's a big separation of technology roles versus people. You know, business roles, I mean, like they are even further apart in the distance. So can we, can, we can. Where does this lead us on the path of organization? Can we elaborate on that?
Mattias Altin:Henrik and I had a whole morning talking about it actually, yeah.
Henrik Kniberg:Want to, or shall I?
Mattias Altin:Well, when GPTs became available but I created two virtual. We have a fantastic team we hired now in Stockholm for hops and we, we, we had a legacy system with a platform. So I created a SQL Sarah and she saw SQL Sarah, sql Sarah, she, she knows everything about our schema and the domain and so on.
Henrik Göthberg:So you pushed what data in there?
Mattias Altin:No, no no, no kind of sensitive data. Was the schema our domain, like what? What are we doing? So I uploaded everything I could and actually work quite well. I can ask Sarah about how should I get this information out? And then we I also created analytics, anna, so we have two team members. That was Tesla. Can we add AI team members to the team? It was more of a fun thing to try the rest of the. My colleagues was like so, I was, but I kind of that was my first dip into that.
Henrik Kniberg:And Henrik, we talked about this today in our yeah, um, yeah, I think what's happening now, what's going to happen more, is that, looking at a development team, is that it really will be your colleague? Yeah, in the sense that now a lot of teams are doing like you mentioned. I have a problem. I'm going to use chat GPT instead of Google to solve it, but not really changing anything about how they work. But I think you know, what's happening now is we're starting to get into AI, integrated into the development environment. So now you can not go to chat GPT, but you're typing something right into your code and they start changing your code in place, which is another level of productivity. But what I think is interesting is when you give it autonomy. So I think what's starting to happen now and I think it's going to explode really fast is the notion of having an autonomous team member.
Henrik Kniberg:So, at your daily standup, or whatever people are talking about, what are we going to do today? And then the AI chimes in and says I put up five PRs just now, the last minute, because I found a few bugs that I think we're, I think I understand what needs to be done. So here's five PRs, please. You know, and if you know, yeah, basically, autonomously fixing bugs is low hanging fruit. And then maybe someone at the standup says listen, this bit of mess we made last release in this part of the code. Do you think you can take a stab at cleaning that up? Sure, I can do that done. There's a peer up. Now anything else you want me to do.
Anders Arpteg:Right that that's going to become normal surprisingly fast, so more autonomous agents that helps up for one to clean up perhaps the code and also come up with issues that you can find in the code and PRs asking Initially, but then, of course, at a later stage, it will be.
Henrik Kniberg:We're at our design meeting. We're having a discussion about something. Ai is listening in the background, being quiet, and then someone says, or then suddenly it chimes in hey guys, wait a sec, have you considered a GDPR for that thing? Oh yeah, tell us more. And then it, so it becomes a really a team member. But that self doesn't change the structure so much.
Henrik Kniberg:But what I think will happen is, if you take agile principles, for example, things like working a cross functional team, shipping, often I think the principles are are staying the same, probably mostly, but the practices, I think they're going to be just turn inside out and upside down. Because why do you need to have a full cross functional team? Normally it's because you need all these different skill sets in order to deliver, but now we have these models that know everything. So I think, just guessing here, that in the future we're gonna have very small teams, one or two people plus an AI model that knows everything. We still need the people there to take responsibility to, you know, to give feedback, to decide the prompts, to own the context. So person plus AI is kind of where I think the magic lies.
Henrik Göthberg:So the new team. The new team becomes maybe. So we've been hypothesizing that we need to. We need to productize the way we work and the teams that we build in the future are product size, you know, and in this perspective then product size can be one or two guys, because the rest is done by the.
Anders Arpteg:We've been speaking before about you know, are the roles. I think in Spotify we had so many more increasingly specialized roles that you had you're an, a Java database specialist or whatever it becomes, you know more and more specialized in the type of role that you have. Perhaps what you're saying is that cross functional teams of the future will not have the need for so many specialized roles, because each person can be increasingly general with the help of AI.
Henrik Kniberg:Yeah, I think what I imagine is and this is just speculation, but also a bit of observation is smaller teams and then more teams, because I think the best companies will be the ones that don't just fire people. They will AI, empower their people and then achieve a lot more. So you have smaller teams but more teams. Each team is one or two people. That can be a lot more general, generalized, because they have the AI. But then you still have specialists, but they're not in any team. There's, let's say, I have a database specialist. He's a specialist. His job is to make sure the different AIs are behaving. He'll go prompt the AIs he'll help, and sometimes a little more human supervision is needed, he'll jump in there. But for day-to-day work they won't need that human specialist. He'll become a bottleneck, so to use the team's apology book kind of terminology.
Anders Arpteg:It would be more enabling teams that have specialized skills and go and help other teams.
Henrik Kniberg:Yeah, I'm not sure if that's a term, but yeah, I think we're moving towards a future where we have these small teams that can do very many different things very fast, and then specialists who come in at specific moments to complement or tweak the AI models. Essentially, so, the job of the human database expert is to cultivate AI database experts. The AI already knows more of database stuff than he does, but he knows more about the context maybe.
Henrik Göthberg:I'm testing an idea We've been looking at DERDAX into thinking about agents. That sparks us. Wait a minute, we can go back in research and we can talk about agent-based modeling. And it's a huge field of research around agent-based modeling and understanding agency and ultimately, hypothesis is that you can understand an enterprise as agents, as different agents. You can model an enterprise or an organization as agents in relations to each other. Now the benefit by going down that route is that now you can start thinking about who is the booster and who is the reach, who is helping someone else to do their job, and in this sense, you get to a slightly different understanding for how we want to organize, which is much more sort of ecosystem. I mean agent-based cybernetics. It's like we're looking at complex system theory kind of stuff. That is sort of fueling ideas. That basically, well, if we look at organization in this way, I can actually put an agent artificially here, but then I need to have these other agents human here. What do you think about that whole? You know to start exploring.
Henrik Kniberg:I think to raise abstract level a little bit. I give some examples of what I think is going to happen, but it's really just guessing. But one thing I am pretty sure of, though, is that things are going to change a lot, so I think the mindset people need to have in the companies is all the knowledge we have about how to organize human work is based on based on assumptions, and we need to question all of them, the assumption, the heuristics, the rules, yeah behind.
Henrik Göthberg:Why do?
Henrik Kniberg:we have departments, why do we have managers? Why do we have companies? Why do we have teams? Everything needs to be revisited.
Henrik Göthberg:Actually, this is what. I you took my topic up one abstraction level. That the rule of thumb. So how to organize? That's what we need to question.
Henrik Kniberg:Yeah and just go and some things may still apply. Some things change and that just means that as a leader, you just need to kind of take a step back and be willing to challenge things a lot more than in the past. And that's going to cause organizational change. You'll need to find a suitable pace, not change too fast, but it is going to change and you got to kind of be ready for it. And whether it's going to change in the direction I said or some completely different direction, I have no idea, but I think the underlying truth, the profound truth you said here there are heuristics, there are rules of thumb that we have learned in university.
Henrik Göthberg:Do that we have learned by experience in our companies? That needs to kind of be questioned. Like we have models, how we think All models are wrong. Some are useful yeah, it's George Box's box statement and now we need to really question which one are now useful still and which one is radically anti patterns.
Henrik Kniberg:Yeah, Because humans are still humans. We're the same. So some things are going to stay the same, but some things are going to be very different.
Anders Arpteg:So that's just time to just move to another topic as well, perhaps connect to this, and that's cool Because hopscom, I think, has as one of the core themes of self improving organizations. Yes, I guess it's related a bit to this, how you organize actually companies as well. So can you just elaborate what you mean? What are really self improving organizations?
Niklas Modig:First of all, it's about improving. I think, if I elaborate a little bit on that first and going back to the Agile discussion here why I'm thinking a lot about how to use or AI discussion. How to use AI, I mean, if we go back since I was in academia for 15 years, we are drilled about defining what is a good theory, and one of the best theories of all time is survival of the fittest and survival. That was what Darwin wanted to explain and he explained it with the animal that is fit and from a theory perspective we talk about, we have a dependent variable that's what we want to explain and we have independent. That's basically what do we do in order to manipulate that one. So fitness, the most fit animal was the one who survived.
Henrik Kniberg:Where they fit enough animal.
Niklas Modig:Exactly. So going back to the lean here and this is a little longer story to come to your answer but I think one of the problems with lean was that they focus too much on being fit and they lost the understanding about well, what are we trying to improve? The survival. So we became. That's why it become a verb, it's become an adge, and then that's what happens with agile as well. Oh, we have agile agile everywhere. So it's all about being fit and we forget a little bit about what is survival.
Henrik Göthberg:What is purpose? What is the?
Niklas Modig:purpose? What is outcome? And I can see here a little bit. Now we're going. We have done that with lean, we're done with agile. Now we go all in AI and it's the same thing. It's nothing about survival, it's the being fit, it's being agile, it's about being AI.
Henrik Göthberg:Agile is ketic A and the I.
Niklas Modig:So what is interesting here? When we talk about agile and productivity, then we add an aspect to it. Well, how can we use AI in order to increase productivity? That's a very interesting question. Or how can we use AI in order to decrease lead time? That's also really interesting. And that's when you, if I ask chat, chat, gpt, how shall we apply lean, then I get a bullshit answer because lean is undefined. It's very, very blurry. But if I ask chat to GPT, how can I decrease the lead time of a process within heart surgery, then I get 20 crisp answers about exactly what type of practices to change in order to decrease lead time within that certain context. And now it starts to become interesting, because then I have the survival plus AI or whatever it is.
Niklas Modig:So, going back to your question, self-improving organization. For me, I think improve it's about improving, that's the survival of a company, and self-improving that is definitely kind of a choice. How do we improve? And I think that, since the world is not so functional nowadays I mean, my wife is from California she's applied for Swedish citizenship and it takes years, and when she moved here it took 22 months to get a visa. So we had to move to Norway, even if we were married.
Niklas Modig:And I did a cancer diagnosis seven years ago. We took six months and I almost died just because I thought that I was going to die, because I have three of my best friends died in cancer. So there's a lot of things that is happening here in the world that is not so great. And going back to what lean is it's about, or agile, it's a system theory. So if we need to improve, then we need to involve the system, and then we can definitely not say self-improve, yes, then we need to involve everyone so they, the system, can self-improve. If we don't involve everyone, then it's impossible to self-improve since we are not involved in the system. So self-improving organizations is organization that can improve without an external support. They have built that capability, the capability, building capability, so to speak.
Henrik Göthberg:So I used, I tested, I used a lingo when we were in Scania that we were. This is about data and AI innovation. It's about taking the financial side of Scania loan and leasing and repositioning as a fintech, because it has to compare to how it worked and we use the work. We need to build the engine, we need to build the muscle in order to understand how we become better at continuously improving or innovating, whatever the direction of the new transport ecosystem that no one really knows how it looks like. For me, you are using a different lingo and semantics to talk about. We need to build the machine that builds the machine, and we came to that.
Henrik Göthberg:I came to exactly the same conclusion. You know, when we talk about data, you know, oh, we have the old school way of transformation. This is assist. This is the to be. This is the gap analysis. How can you do a gap analysis on a completely galloping target in terms of productivity frontier? You can't. You can only build the machine that can build, the machine that can adapt. And all of a sudden now we're moving in a paradigm where we go from economies of scale and efficiency to economies of learning and adaptability. And this is the core topic how do we build adaptable companies? I think that's what you're talking about, right?
Henrik Kniberg:But perhaps also we have a word, oh yeah, it's time for AI News Brought to you by AIW Podcast.
Anders Arpteg:Awesome. So we have this middle section in the middle of the podcast where we take a small rake not always that small, but we aim to make it small where we just speak about some of the recent news and then we get back to the discussion we just had.
Henrik Göthberg:So it came about when he was ridiculously moving fast in terms of new things being released in the world of AI by the minute, even the samultimate debacle by the minute. You know what is this? Yeah.
Anders Arpteg:So who wants to go first? We have a couple of huge stories, of course, happening, but I'll ask the guests first, I guess Do you have a story that you want to start with, or should we take one, henrik, me or Goren?
Mattias Altin:I think the funny thing today was the Elon Musk open AI fight.
Anders Arpteg:You want to open that, or should I?
Mattias Altin:Yeah, well, it was. You stole my news piece. Let's go here.
Henrik Kniberg:No, no, you do it. I prefer you to do it. Cage fight, Not this time. I start you finish.
Mattias Altin:Yeah, no, the whole thing about the lawsuit. And Elon's response was basically if you change your name, I will drop the lawsuit and change it to closed AI. But let's give you a bit of a background to this as well.
Anders Arpteg:So it started with Elon Musk suing open AI and Elon Musk actually was one of the co-founders of a open AI to begin with in 2015 or something, and he left in 2018, I believe because he thought the company was moving in the wrong direction and becoming increasingly for profit. And now he's suing the company for saying it is not developing AGI for the good old mankind. Instead, it's moving to a for-profit organization as a subsidiary, basically, of Microsoft. And this is kind of interesting because then open AI came back and said oh well, we're going to disclose all the email to send to us, elon, where you said well, you should go for profit because that's the only way you really can develop AGI.
Anders Arpteg:And there were a lot of letters where Ilya Sutskiver, which was one of the co-founders as well, and other people Greg Brockman, etc. Said they spoke to him and said, basically, they should go for profit. And now Elon is suing them for going for profit when he said they should. I mean, it's very much a double standard in some sense, at least when you listen to what we've seen so far. But I think what you said, mattias, later, is the best tweet of all, or X of all. Can you just say it again?
Anders Arpteg:It was such a good thing, I just thought on the way over here.
Mattias Altin:I thought it was funny. Basically in that response, elon was basically if you change your name to close AI, I will drop the house, so basically you're not open anymore, yeah we can give him that he has humor.
Henrik Kniberg:I did a little bit of a.
Henrik Göthberg:I thought it was interesting. So I someone on YouTube obviously had found the document. What is the lawsuit all about? And then like a 45 minute breakdown read through, read review of the actual lawsuit.
Henrik Göthberg:So what is it actually that Elon is suing for in detail, what is his objective or strategy here? And you kind of see it's like a couple of different main things. So, first of all, obviously trying to argue that I paid money and I put $44 million or whatever it is into a non profit approach and therefore the investment I made has now been used for a different purpose. So this is one angle. I'm not saying that is double standards, for sure I get that.
Henrik Göthberg:Second one is saying, which is very interesting, he is highlighting in the claim in the lawsuit is sort of trying to highlight that well, when you now have licensed, what is it that you have licensed? You have licensed what you had which is not AGI. So he's sort of trying to put the wedge in and say whatever AGI is coming out of opening up potentially is not part of the licensing. So there's nuances here. And then one of the final topics which is quite sort of remarkable is that he refers to I don't know if you know what this is all about, but he refers to what was leaked as the Q-star approaches and to even put Q-stories about now we're talking about the developments in opening AIs. That presumably has something to do with much more, much more sophisticated, much closer to AGI than we think about. But it's like quite remarkable that in a formal legal claim he's referring to Q-star, especially since it's just a rumor, right.
Henrik Göthberg:Yeah so maybe that puts some fuel to the fire of that rumor, doesn't?
Anders Arpteg:it. But he also says that his own approach, the XAI thing that he's building that had the grok with a K, a chatbot in it that is deployed on Twitter or X, is supposed to be open.
Niklas Modig:But it's not.
Anders Arpteg:I don't see any open size on that. I don't see any open model on that. How is that open? But I think, if I were to try to steelman his argument, what he's really trying to say is that he wants AGI to be available for people and not locked into one company. But in what ways is XAI that it's locked into Twitter or X and open AI? Then potentially it's locked into Microsoft?
Henrik Göthberg:yes, I think this is more a strategic move. I think it's a way to divide and conquer, or I don't know. What do you think? We don't know anyone else, of course, but what?
Henrik Kniberg:is your take on that? I used to think of Elon as a very rational person and an inspiring leader that changed every bot X, and now I think he was a very irrational person, so I don't really try to make any sense of what he does Exactly, that's a good one.
Anders Arpteg:Well, if there's nothing else, it's good entertainment for all of us. This is a better than any SOPA bra we have had three years, what else? Any other ones.
Henrik Göthberg:Do you have someone If you don't know?
Henrik Kniberg:I guess it's. I just kind of think it's interesting, with the latest cloud cloud models, yeah, that basically GPT-4 has been the king for quite a while in terms of capabilities. Nobody has really gotten near in terms of what GPT can do as a general model, gpt-4 that is. But now I haven't used cloud myself, but I've seen enough data from other people to indicate that yeah, seems like GPT-4 now has a proper competitor.
Henrik Göthberg:Can we contextualize cloud? What is it? Who is it?
Henrik Kniberg:Cloud is another model and I believe it was a breakout group of people from OpenAI that broke out and created their own model from Google and from the front.
Anders Arpteg:Yeah, yeah, helped me out with the data.
Henrik Göthberg:Silicon Valley yeah.
Henrik Kniberg:So it's basically an alternative to, and it hasn't been as capable. It has had its own strengths, but in terms of as a generically useful and capable model, gpt-4 has been quite alone on the throne for quite a while. But now it seems that there's leapfrogging happening, which I think is good. So now they've released three models with different levels of capability. That's also good because you don't always need the Rolls-Royce super expensive model for everything. But they released three models and the most expensive, fancy one seems to actually be a proper competitor, maybe even exceed GPT-4 and other ones. But yet you can also choose to use cheaper, faster model when you don't need to have, like, a Rolls-Royce. So I think I think it's good that GPT and OpenAI isn't alone on the throne.
Anders Arpteg:It's really good because the concentration of power, I think, is one of the most dangerous thing that we can have in the future, where we potentially have AGI and the AI divide that we have spoken about so many times. So it is really impressive. It seems still to be lagging potentially on the addition part, where Gemini is still in the lead, but from a from a texture point of view, to actually break and be taking lead of chat GPT-4 is really impressive and really good that someone is competing with them. But that makes you also think you know how long time will it take before until OpenAI come up with the next model.
Henrik Kniberg:So I think it's just going to be leapfrogging right. Every model is going to be oh my God, they just beat this model and then yeah, it's a bit more expensive than OpenAI GPT-4. Yeah, about twice, yeah, almost.
Henrik Göthberg:Yeah, 50%, all right. And now the news. I want to do a tidbit, a small one, and I'm going to lead into it, but of course it's an understope. There's an archive X publication that came out this week and it's highlighted and it's called the ear of the one bit LLMs. All large language models today are in 1.58 bits or potentially a lot more energy efficient and cost efficient we are now talking about. There will be so many evolutions on how we think about and doing this. We are so far away from an end state in any of the stuff we're doing and here we now have. You know how can we make it a one bit approach? Could you help us, anders?
Anders Arpteg:What is it I think you spoke about when I was in here for last week, but used a little bit, but I think it's worthwhile.
Henrik Kniberg:But what is the one bit approach? I don't understand yeah.
Anders Arpteg:OK. So basically, instead of having the traditional FP32 or 32 bit floating points or going to FP16 that a lot of models are using in mixed mode, going to FP8 only using eight bits in floating point, or even using eight bits for integer mode, which is even faster than the floating points, and there are even, like four bit integer models that are just using, you know, integers instead of floating points. Now they have a one bit model and the really weird thing with this is not really one bit, because it had minus one, zero or plus one. These are the only three possible values that a single neuron can take, or soon up, since you were to be correct. So the parameter can take minus one, zero one. So if you take a log of three, it basically is 1.58 bits.
Anders Arpteg:Anyway, the really strange thing with this is that they achieve the same level of accuracy as FP16. So using only 1.58 bits, they achieve the same performance as 16 bits, and then significantly lower amounts of memory and significantly faster from a time perspective. This is like any, any any negatives?
Henrik Kniberg:Was it only positives?
Anders Arpteg:I haven't seen any negatives and you know. One other positive thing is that when you go to this kind of minus one, zero one, when you do a matrix multiplication which is basically what all these kind of large models do multiply the vector with another matrix, they move from actually having to do multiplications to only additions. So you can just add up the numbers you don't need to multiply, and that significantly reduce the complexity of the computation.
Anders Arpteg:So it's significantly faster and then achieving the same performance. It's like magic. It must be something wrong with it, but yeah, kind of like what's the catch.
Henrik Göthberg:But this is interesting now because we're having innovation in the technology on so many different abstraction levels. So now we were talking one bit. We are going really hardcore down into how we understand what a neural network is and how you can load that into bits. And that has, of course, huge implications because all of a sudden now that means that you are very much closer to getting something on the edge, like we are talking about Gemini, right? We're talking about how we have the large language model and we have the one for the phone, and we are getting very close now to where we can have a large language model driven operating system or something like this, which is once again completely rethinking how you code right Completely.
Anders Arpteg:And it's also, if we connect back to the previous story and or the first story with Elon Musk suing OpenAI, what they argued a bit in the beginning was that they only asked for like a hundred million dollars and Elon Musk said you need to please raise a billion dollar. Basically, the compute is too expensive. So I mean the personal cost is not the thing that. The thing that really makes the AGI approach really expensive is the compute. So anything you can do to really reduce the compute expenses that you have is super important and also to make these kind of large more available for non tech giants, the super scalers and then take the CO2.
Henrik Göthberg:Take the whole energy. This is going to be a massive problem when we think about what this is going, of course.
Anders Arpteg:I think this will be a trend for 2024. We will see the models not just growing and growing, they will continue to grow of course, but the innovation happens on many abstraction levels but many different problems.
Henrik Kniberg:Yeah, I'm not like some people ask me because of my background with climate change stuff, like you know. What about the energy costs? Yeah, which is, of course, a major issue, but I'm actually not so worried about it because this is where all the interests are aligned. Everybody wants to get the costs down. They're expensive, they're slow, so all they can, all our systems are aligned towards bringing the cost down, so I think they're going to keep coming down over time.
Henrik Göthberg:Yes, in the end, what do we call it? Is it more slow, or whatever? It's going to apply here too.
Henrik Kniberg:But of course I have to factor in that, sure, the energy cost is going down but the usage is going up.
Henrik Göthberg:So you know it is for the near term going to be a problem. It is an interesting one, right, and it's a little bit like the big politics down to a little bit the one bit problem. Interesting let's. Do we have another one or do we? Do you want to add? Did you have anything? You don't have a mic today.
Niklas Modig:I have a question to you guys. Yeah, what is the best AI news for a person that doesn't know anything about AI?
Henrik Göthberg:This one is because of podcast.
Niklas Modig:No, but understand, I didn't know. I'm like everyone else, I don't know so much about AI, but until chat GPT, that was the first real news for a common person. Like, hey, I can start using this and now I can even show my friends and even if I'm super beginner compared with you guys, I can still show off for my dad or for some friends. That was a new, that was news. What is the next thing that will change my behavior as an amateur? What?
Anders Arpteg:would you say? I think one thing that I usually speak about when I give public talks is and Sam Oldman say the same thing, google say the same thing, but I'm going to repeat it anyway because I think it is very profound and it's basically is connected to the story we had before. You know how do you do self improvement in a company, and any person or company that is using AI will become significantly more productive and they will overtake other people or company that are not using AI. So the AI should be seen as a significant productivity boost for a person or for a company. It's not a replacement for a person or a company, but it is a significant productivity gain that will overtake other people and companies that do not use AI. So I think any kind of leader manager in any company that does not realize this is just a matter of time before they will be overtaken by other companies.
Henrik Göthberg:But to answer your question, how will the normal person? Because there was like we've been talking about this for years and we understood more a little bit like shit. We got this a whole moment that a lot of the media put out there to the most people. I had that a whole moment a couple of years earlier. Right, but of course I couldn't. What happened with chat GPD was profound and your question now is when is the next thing and how will we as commoners if you want to, I don't know that's not a good word, but how will? People are not involved in the industry experience. The next it will be a next bump. What is the next one?
Niklas Modig:When will the regular person change their behavior due to.
Henrik Göthberg:AI. I have one, but I think.
Henrik Göthberg:I think. I think I think there's a couple of things to think about here. What you have experienced now is that we've been trying to improve in organizations for a year, but it doesn't really fly. What happened with chat GPD is they create a user interface that allows you, as an individual, to play around and basically right now, at your fingertips, we can increase our productivity as individuals, as persons. The next problem is, of course, that in order to have real productivity gains, you need to take that individual productivity gain and put them into systems or systematically as organizations, and now we see the first movements on this. That is sort of hacks. It's really hacks. It's the rag.
Henrik Göthberg:You know how people do this stuff that you can do individually, and they kind of fix it a little bit. They haven't changed the underlying process, they haven't changed the underlying operating model. They are sort of they are sugarcoating what you're doing with AI. So so one big leap now which will not be a leap in the same way, will be sort of moving into these systemized approaches where it's not me as an individual only fundamentally getting in productivity gains, but it's the system in the organization. And now, what is it that we talk about then Then I think it's the era of AI autonomous agents.
Henrik Göthberg:So, basically, now we have a transformer model or something that can help you automate individual tasks, one simple, small task. The next problem is, of course, we want to give a little bit more abstract problem task that the AI needs to plan and sequence. So the simple task like I want to order pizza and that is actually if you're going to do that as an agent, it needs to go out on the Internet, it needs to find the pizza, it needs to order it, it needs to secure that it gets so. So this is a sequence of steps that the AI does.
Anders Arpteg:So the next major, profound aha will be when it's a little bit more general and we will call that I think it would be really dangerous to tell companies that are not in AI today to say, start developing autonomous agents.
Henrik Göthberg:That would be a horribly bad idea. Scary, scary idea. So start by.
Anders Arpteg:Why? Because it will fail. It will be like the original chatbots of customer support from 10 years back, when I thought that you do it and it will fail.
Henrik Göthberg:What I'm just saying is you can't jump the gun.
Anders Arpteg:You need to learn by doing the main thing that we can do as a company or person is to get enhanced, is to augment ourselves, and that is the best way to start using AI, and anyone should not do anything else.
Henrik Göthberg:Get your hands dirty.
Anders Arpteg:Get that done, probably first.
Henrik Göthberg:Get your hands dirty first, and then, when you get your hands dirty, it will evolve and the evolution will be faster than we think. But so, but, but. What was remarkable with chatGPT was that it was like from a mainstream point of view, it went from one plateau and it just went to the next plateau. What is the next plateau? Will we see that again? Maybe around autonomous agents? Maybe now we don't know what chatGPT is going for. Are they having two stars around the corner? Is it a major leap now in that you can use give?
Anders Arpteg:Let's get more philosophical perhaps in the end and we have so many more topics we'd like to ask.
Henrik Kniberg:I just just add my quick one the next show off moment it's going to be you're in the couch, I'll do that so I can look at you. Hello, you're in the couch. You put on Netflix or whatever, and you ask your, your father, your friend like what series do you like? I kind of like Breaking Bad. Right, cool, would you like to see something like Breaking Bad? Yeah, yeah, would you like anything different? Or do you think that was perfect? I'd like to have a bit more humor in it. Ok, go Prompt it.
Henrik Göthberg:And.
Henrik Kniberg:I'll just generates it in your face and then you're stuck in a sofa forever and you all starve when we all die because we're watching too good binge watching. But just looking at Sora, the latest video generation. It's just one minute, but I think that's going to be one of the big thing that affects normal people the ability to generate content instantly.
Mattias Altin:My take is that we talk about our Hubs project, how we're going to incorporate AI, and we talk about the visible AI that you won't notice it. It will be there in the background. It will be very powerful, but you will not feel intrude. It will be something that just enhance your whatever you're doing. I'm more of a yeah, I think. What's your thoughts on Rabbit hour one?
Henrik Göthberg:Yeah, we are going to talk about it. Ok, sorry, I'm doing your punch last night, no, no, no, that was not a news, but it was I was leaning into that and we'll try to understand what's the next big thing. And here we have an AI phone or LLM phone.
Mattias Altin:Yeah, did you order one, Not yet have you. Yeah, you ordered it.
Henrik Göthberg:I couldn't help it. Sorry, you couldn't help it. No, I'm a geek, I need it. Yeah, we can close the news topic, you think, and we move into, because we already moved on. I think, but you were referring maybe for the listeners who doesn't know what we're talking about, you were referring to the rabbit, one rabbit or one, I think. So, yeah, which? What is that? I think you can do better job.
Henrik Göthberg:I think it's a hard time explaining it, but it's a phone that is not a phone that is really driven by a large language model and you kind of stick it on top of your own phone and someone explained it to me. I have all these stupid apps now and we get and bombarded. We have hundreds of apps. So you want your agent that basically you can say, order a pizza, and then he will figure out which app to connect to and do the work for you or whatever it is. So it's sort of you, it's like a chat, it's like an agent to chat, like it's like a device right that you can chat and we'll see. How did you understand it? I actually, I just you ordered it, so you just wanted a cool coffee.
Mattias Altin:I'm a sucker for cool gadgets, but I just watched the demo and think, for that kid, this might be the next big thing that Nicholas asked for and I just want to try it early, but I don't really understand how it would work for my daily life. Yet Standing talking to my phone in the street.
Anders Arpteg:I don't know, we'll see. Good thing, of course, is the change in interface that we will see now Both the programmers, but also for the rabbit one to just speak to something and make it do stuff. It's like the fake video from Google for Duplex back in 2019 or something. If you saw that and you could tell it to just order me or make it a booking for my hairdresser and find a good time for me and it was faked but still it was a cool idea. Now we write one. It's supposed you just can speak to it and say, please order a Uber or taxi for me and we'll do the work for you and get it done. And of course, that's a big change and it's similar to programming, I guess. I mean, some people say that the future programming language will be English, and that's already true today to the large extent, and this kind of change in interface and the way that we work with machines will change a lot in coming years.
Henrik Kniberg:Sometimes you don't want to program in English, sometimes you want to program in code because you want predictable results. So I think there's going to be space for both.
Anders Arpteg:Yes, I agree.
Niklas Modig:If you want creativity and intelligence, then use a prompt.
Henrik Kniberg:If you want predictable results, write code or make a prompt that writes a code for you. Exactly.
Anders Arpteg:Yes, combined the two.
Mattias Altin:Yeah, I remember back in 95, around the time I also started, that we talked about the next generation languages. Will just you don't kind of write code anymore, it's the dragon drop and all that, but it never really took off like level four, three, four languages.
Anders Arpteg:I love to get into the awesome generative AI video that was going viral like crazy as well. Perhaps you can, henrik, just describe it. How did you, what was the thinking to create it? And, yeah, how do you do these kind of videos? I'd like to just understand how you go about actually making a video like this Right.
Henrik Kniberg:So I guess that's two questions maybe. So maybe in this particular context of that video, I hesitated a little bit because there's so much information overload right now around AI, so I wasn't initially planning to make one of what can I add to all this noise. But then at some point I felt when I talk to people, when I'm doing presentations or workshops, I noticed that people are thirsty for a high level summary because there's so much information that they get more confused by all the information. So after I just decided, I went back to my summer cottage blocked off a week, essentially and just did say that I'm going to put together another video that just makes this clear.
Henrik Göthberg:How do you make it? Do you make it yourself or do you have someone who makes it? Into what programming? What applications are you using practically?
Henrik Kniberg:It's handcrafted. That's the irony in it. I made this AI video, but I mean it's all handcrafted. But I should draw on a tablet and record my screen.
Henrik Göthberg:Actually so, because I think there are sort of ways of doing that, but this is what you're looking in your videos. You have drawn it.
Henrik Kniberg:Yeah, it takes forever, it's a very nice handwriting. And it takes a long time. And I'm not as good at drawing as the video might make it seem, so I have to do a lot of retakes.
Henrik Göthberg:And then you'd stop motion on the different no.
Henrik Kniberg:I just record and I draw and I'm like, ah crap, Control-Z, Ah crap, Control-Z. And then after 15 Control-Z, it's like, oh, I managed to write Einstein sitting in a chair, and then good, and then I just saved that, but then you can draw.
Henrik Göthberg:I mean, henrik, you can draw, because your small little drawings they are nice.
Henrik Kniberg:Oh yeah, yeah, no, it's just, that a real artist would do that instantly. I need a bunch of. I need to do it slowly and with retakes, so I don't want to just be humble.
Niklas Modig:He's an amazing artist. He's very humble.
Henrik Göthberg:How long did it take to actually make that video.
Henrik Kniberg:I actually tracked that time a little bit. It took about 60 hours 60 hours, but a large part of that was figuring out, like obsessing over there, every word I'm going to say, exactly Because it's compact. It's a whole day course essentially in 18 minutes. It's crazy, yeah. So I spent a long time obsessing, counting every word, so I know how many words I speak per second. I want the video to be maxed 15 minutes. I failed it. It became 18 minutes, but my goal was 15 minutes and I spent probably two days working like two effective days, which corresponds to our normal work week.
Henrik Göthberg:Right, and then you have the manuscript first.
Henrik Kniberg:Yeah, so I obsess over every word and I count them until I get to a point where I got my message across with the fewest possible number of words.
Henrik Göthberg:Yes, Word count per minute. Yeah, and now you then think about what are the visualizations that will slow naturally with this.
Henrik Kniberg:Yeah, I grab various, typically snippets from my various presentations. So when I talk about this, I'll probably show this thing. When I talk about that, I'll probably show that sometimes. So I put it in a slide deck actually the manuscript, and so I cut it. I do some clip art. Sometimes I ask GPT to generate something, sometimes I draw a little sketch. It's super ugly, but each page in this presentation is essentially what I call a chapter, which is about one minute of content.
Henrik Göthberg:Yeah, because you can feel it, because when you're trying to break down watching the video with a nerdy eye, then you feel the chapter.
Henrik Kniberg:There's a structure. Yeah, so every chapter is on the manuscript. There's one is a slide takes on the left side and just a bunch of sketches on the right side. But then I know what am I going to say Roughly, what am I going to draw when I'm saying it, and that's about half the time just doing that. Yeah, To get the whole flow. Yeah.
Henrik Göthberg:This is a little bit like the storyboarding.
Henrik Kniberg:Yeah, but it only works because I've done a presentation 40 times before. Yeah, practicing with the audience, so I already have all the. I know what works.
Henrik Göthberg:Yeah, so the background. The background then starts way earlier, because if you 60 hours is bullshit, 60 hours is only the production of it. This is the production of the actual stuff, but because you have groomed it. What is this really? A one day generative you know? Have you sort of made consultant money on that sort of you know what's the background?
Henrik Kniberg:of the content, but you have given a lot of lectures. I've given lectures, talks and a very lots of different contexts, written articles and the whole thing, or typically parts of it. Parts of it, yeah. So this is putting it all together, yeah.
Anders Arpteg:Cool and for people that's interested in what we're actually speaking about, they can search for generative AI. In a nutshell, yeah, right, yeah. It's amazing if anyone doesn't have an understanding of generative AI. I would really recommend this is probably one of the most technical videos.
Henrik Göthberg:The first comprised 18 minutes on the landscape.
Anders Arpteg:I will try to make or ask Chatterbt to make a pedagogical video about X in the style of Henrik Nieberg.
Henrik Kniberg:This is actually I want to talk a bit about this because this is funny. People ask me how do they use AI in this? I use it for basically only two things in this video. One was fact check, because I prompt the GPT and says I want this to be factually correct but also very simple for normal people to understand. So I can't oversimplify. But I also want I do want to simplify, but not oversimplify because then it gets too incorrect.
Henrik Kniberg:So it helped me, say when I said will you use metaphor? This metaphor here is very useful, but it's also incorrect, like too incorrect. That was one thing it helped me with. But the second thing was I prompted GPT and I said the most important thing for this content is I want I naively want to reach both the experts and the beginners. That's normally not possible, but I said I want to reach both beginners and experts. Probably failed, but I'm going to try. And GPT was very helpful there because it told me you'll lose the experts when you simplify this much and it said you'll lose the beginners when you go this advanced.
Henrik Göthberg:So you had that as framing problem so it will sort of help you balance.
Henrik Kniberg:So simplify a little bit more or make it a little. So I added some things that are just for the experts. I showed an example of a research paper I popped up just for a few seconds to emphasize one point that I was making, because I know that the expert would realize that yeah, okay, henry knew about that paper. I'm not making stuff so little things just to make sure that I'm casual.
Henrik Kniberg:And the funny thing is now, when I read the comments, I feel very happy because I get comments from people like I'm a teacher at a grade school and I don't know anything about AI. And now I get it and I'm teaching my kids and my class and they get it. Thank you very much. I'm going to spread this to all teachers and that just warrants my heart. And then the next message is I'm a university professor in machine learning and I'm teaching students and it's like can I use this in my course?
Henrik Göthberg:I'm like yes, I got both Wow. I don't think you could have done that without helping. What satisfaction Cool.
Anders Arpteg:Awesome and the time is flying away here, and I think we've spoken a lot about organizational changes that AI will bring and how to do that in a lean, magi way, et cetera. But perhaps one aspect that we haven't spoken about is more how will it change their products? I'm not sure if that's something you are focusing on, but one thing that we have seen at least the tech giants doing is simply add a language model to each product they have, to co-pilot to Windows, to the web browser, to the Office suite, to Workspace, in Google, adobe, added to all their products. It's simply like an assistant or a co-pilot to whatever product they have. Is that something that you think is a good way? Or see that AI will be used in the future? Or how do you see product changes, or how do people think about product changes in the world where we have the AI to help us?
Henrik Göthberg:Like.
Mattias Altin:I mentioned earlier, I think we're talking about invisible AI. Yes, you said that it should be something that is in the background. It's like a butler in the corner of a dark room that come in and help you and disappear, and hopefully, what we're building now.
Henrik Göthberg:We will be able to utilize this how are you thinking about implementing AI in your TMS? Have you thought about it yet?
Mattias Altin:We're kind of talking about it. We haven't started doing it yet, but it is in that context.
Henrik Göthberg:Can you build a butler, a TMS butler?
Mattias Altin:We have two leading persons within their field. Wouldn't it be great to have these people?
Henrik Göthberg:Exactly Avatar Henry Neymar.
Mattias Altin:At your table virtually, but not being annoying.
Henrik Kniberg:You're not being annoying, but something that's there to kind of notch you in the right direction and also maybe find patterns, simple things like if a customer is using our platform, they're consuming a bunch of content, they're inserting lots of improvement ideas, things they want to improve. That's a lot of data. And then if our system with knowledge, plus all the knowledge that language models already have, we can provide some tips like, hey, this problem that you're having, would you like to have some ideas for how you can solve it? And then so it can provide just ideas.
Henrik Göthberg:You can imagine. I think this is some of the problems with learning management systems or content management systems how the fuck do you organize the information, what people have different routes, how they want to go through the material. We have different use of personas. What's the journey we are following here? So one way to do that is basically can we make it, can we guide it, Can we make it more conversational in style of learning? Can we have a knowledge craft underneath? Because the way you can, when you talk about the different dimensions of becoming data with AI, ready organization practices, team leadership there are so many tangents and different roles has different problems. And then you have the temporal problem. I'm early in the project, I'm late in the project, I'm going down in the ditch, so you can imagine there are so many ways that you could want to question and prompt all the content.
Niklas Modig:I think it's interesting to think about how can you use AI in a product without taking away self-fulfillment from human individuals? Because I mean, going back to the example with Saab, we were able to ask a question to 1,500 people at the same time and if all those 1,500 people answer a question and they come up with three IDs each, that's 4,500 IDs. From a cognitive perspective, it is when I come up with something. That is the process when you create engagement and ownership. But if my wife says, hey, niklas don't eat the Sembla, I say what about you? You don't eat the peanut butter? Then I mean it's like I don't want to get a tip from her. But if I say, hey, babe, I'm not going to eat Semblas and I feel so proud of myself.
Niklas Modig:So it's the process of coming up with the answer. So, going back to this 4,500 IDs, I don't think AI should be asked what shall we improve? And then we could say, hey, we have 10,000 IDs in one second. Let's skip the DVDs because you're so slow and stupid. Ai is not. But if we could take 5,500 IDs that people have ownership and say, hey, can you just categorize this and package this in a way so we can see, what are we collectively thinking that we should improve? Then we have a product that AI has packaged for us, but we still have ownership, like I was part of that.
Henrik Göthberg:But this is profound because it goes all the way down to the first question. When using AI, you need to answer what is the objective function? What are you trying to achieve? Going back to fitness versus survival, and if we really understand that, then we can answer what AI should be used for and what it should not be used for. And I think we can make fuck this up big time if we're not careful now.
Niklas Modig:Yeah, I mean, if I don't have an ownership of improving this idea, it doesn't matter. I know that I shouldn't eat Semblas.
Henrik Göthberg:So if that's the first principle, then what can AI do? But if you lose the first principle, you lost the war.
Henrik Kniberg:Exactly, and this ties in with what has become somewhat of a mantra for me, which is that AI and humans together that's where the magic lies. Yes, so if we completely outsource the AI to just run the improvement, nothing is going to happen really. It's not going to have the context, it's not going to have the buy-in.
Henrik Göthberg:I think the ownership topic.
Niklas Modig:And if we?
Henrik Kniberg:just let the humans run the process. Well, we're kind of slow and we don't know a lot of stuff, but if you combine it, so people are driving the change themselves, but their AI support it, so then they can get a spontaneously generated video as inspiring as a Spotify model video, adapt to exactly the problem they're trying to solve. Now it's like you own the problem, but here's a bunch of inspiration and then maybe we can get that magic AI supporting humans.
Anders Arpteg:I think that's the way to go, so to speak. So I think everyone should be encouraged to try to do so Time is flying away and it's time for the final question.
Henrik Göthberg:You think I mean, like we always, we like also to have a bit of a conversation on more macro philosophical questions. We have already kind of touched them a little bit, but I think, because we've been like a big question, we talked about what is the future of organization. Literally, we actually went there. I love that. But if we kind of go from there and go even further out in the stratosphere, what are the key topics? We would like to ask these guys and do you have a specific one.
Anders Arpteg:No, I like to think the normal kind of stuff.
Henrik Göthberg:I don't know if you, if for people who have watched the pod, we have like a standard ending question that it's like you're doing a research under. You know you're building up, we're conducting an empirical research, we have an empirical research going on, but let's do one more question and then go there. And my idea here was a little bit like I think in this at this table and I was thinking a lot with on Matias. Actually, I think there is a contrast on how different companies, depending on their legacy, have stepped into agile and now AI, focusing on the fitness rather than the survival, and I have definitely seen a difference talking to a digital native, or if you said, are you the team to agile and you they were there. It was used about improving.
Henrik Göthberg:And then I have worked with a lot of different companies with a very successful and proud analog history with a lot of dogma I could call it, but I could also call it learnings that made them successful, how they should organize. So what could we sort of summarize or give our sort of pet peeves? You know, where do we go wrong in organization or in this topic and what is sort of the first principles we go right? So like sort of condensing down the conversation we've had today and I was a little bit like trying to contrast. You know, I've seen this work here. I've seen in all the traditional companies they're kind of struggling with this, not naming names and all that. But could we kind of go there, because I think there's a vast experience around the table here.
Anders Arpteg:So what's the question really?
Henrik Göthberg:You know, what are the what, what? What can we contrast what you know, what some of the digital natives have done and maybe what some of the more traditional organizations? They don't have that background, so maybe they are thinking right or wrong about this. So I'm trying to find out the sort of the major blockers and pitfalls that we see where people go wrong and the major principles. That is actually important.
Niklas Modig:And there's a little bit like a summary.
Henrik Göthberg:Super big question. Super big question. I said that macro question.
Mattias Altin:I think if you take like a big, like Spotify to other companies, don't come from there. It's the thinking of everyone needs to start doing something in the digital space. But if you come from a non-digital, natives side, you would think project. It's like a I have a requirement I'm going to lift a certain date.
Henrik Göthberg:You got me, so you are saying product over project. Please, exactly Like one, this is one.
Mattias Altin:And that that most people who work in those organizations that have some had a software product mindset. That's a shift you need to focus on, and once you're going to get thinking like that, then you can probably start counting up with the different native.
Henrik Göthberg:Because that becomes the first principle. If we haven't left the project process type era and go on product, none of the organizational logics makes sense and you would never be done this Like it's an infinite game.
Mattias Altin:You need to continuously improve your product. Understand what your customers and people you're serving needs. Okay.
Henrik Göthberg:So I think this is what I wanted. So please, can we go from project to product thinking more? Okay, this is your view. Do we have any others like this? I think you said it before focus on the survival rather than the fitness. I think that's a profound summary. Like folks from the objective.
Niklas Modig:Yeah, I think it's when something becomes really hype, we over focus on it. I don't know how many hundred times I've gotten the question. Nicholas, our company are having two big initiatives it's digitalization and lean. Which one is the best one?
Mattias Altin:And what's your answer? Lean, of course, no, no no, it's host.
Niklas Modig:I mean, this is just that we dig into hypes. First it was lean and then it was agile, Then it was digitalization, then we have OKR and now it's AI Hallelujah. I mean, I wish that I go back in time and just take away the word lean and say okay, here we have an area where we talk about short lead types. Add knowledge to how to short lead types or create amazingly safe factories. This is the most safe factory that we have ever created. What do we do so? What are the things that we want to achieve?
Henrik Göthberg:But this rant is beautiful because it's really. We're going into hype cycles on the fitness topic of fashion, rather than being precise to what we really want to do, I mean my best story is when I was working with a hospital, the fertilization clinic.
Niklas Modig:You know where couples come and they can't get pregnant, and they had like 40 different KPIs that they were measuring with the quality of sperms, amount of sperms, quality of eggs, amount of eggs and so on. And I mean if I go there and with my wife and we are not able to get pregnant, I don't really care if I have slow sperms, it's not really what I care about. I want to get pregnant and that's what they come up with. So they changed all those 40 KPIs and they say we have one KPI and that is baby take home rate and meaning which couples have a baby to take home. That's the only thing that drives customer satisfaction. So I think with every company should ask what is our baby take home rate and how can AI help us increasing or decreasing that?
Anders Arpteg:I mean, that's what it's all about. I guess it's simpler to Elon Musk. You know, in his SpaceX mission he says you know, every decision I take, if I should change the screw in this way, change the engine in that way, is being decided based on will it get me faster to Mars or not.
Henrik Kniberg:So, thinking about the great grand vision in the real goal that you have, that is really what you should make every decision With Tesla right With the time to eliminate the last fossil car, right Also fuel car, all oriented around that. I think it's super motivating when you find those metrics.
Henrik Göthberg:But here we have two first principles Project or product as a fundamental principle to think about why we organize wrong. Focus on the core objective, get rid of all the fucking fluff and go and be more precise on the objective. And now, Andri, do you want to add anything to that? I wish I had just one, but I got three. I'll give you three, you can pick one. How about that, that's beautiful.
Henrik Kniberg:So these are the things that I keep rediscovering as, yeah, this is really important. And one is experiment, just as a company experiment with your product. That seems to be just a key thing, regardless of everything Experimental culture. The second is visualization, visual management Make stuff visible, whatever it is. People just get smarter and even AI models become smarter when they see stuff. So make things visible, especially the things that matter. What's between us and increased bayou takeover? Yeah, exactly. So, yeah, I'll explain that. You give buzzwords, perfect.
Henrik Kniberg:And the third one is also stolen from you, the respected people. But like, at the end of the day, people that are motivated and happy do a better job, and even with AI models all over the place, it's still people in the loop.
Henrik Göthberg:Yeah, it's not interesting. We talk about tech, tech, tech, tech, tech, tech and no one seems to reflect this is a social, technical paradigm. It's people using tech, it's the value in use we want to get out of it. And all of a sudden we talk about tech, tech, tech, agile practices, that this is a technology product. You know where's the people?
Henrik Kniberg:Yeah, and if they're not inspired, if they don't care about your KPI or your vision, then the tech is not going to be okay, so experimentation, people and visualization, visualization.
Anders Arpteg:So favorite leader, favorite CEO of a company that you admire, and I think this is a role model for how someone should lead a company.
Mattias Altin:And why is it Nicholas?
Niklas Modig:Nicholas, he's feeling there Before you can think about that but I have to comment what you said about respect and also about the values I mean. One reason why I love Hendrik is that before we really knew each other, I was working with the oil company. They really wanted us to help them to become more efficient. And they said do you know Hendrik Nieberg? And I was like, yeah, of course I know him. I really didn't. I had his number. So I call him and say, hendrik, you know this oil company. They want us to help them. And he said what type of company? Oh, they're Norwegian oil company. And he said, yeah, yeah, I can help them for free. And I said you can, yeah, if they put their company into bankruptcy. I don't work with that type of companies. Click, I didn't click, but yeah.
Henrik Kniberg:No, no, no, but you actually said that and I think that is one of my.
Niklas Modig:I mean one of the situations where I loved you because being driven by true values. I think we need to bring that in, because how do we use AI without the human values?
Henrik Göthberg:Yeah, that's well said Awesome. So this was kind of a little bit of a wrap up I was going for you gave it a goosebumps moment. Thank you very much.
Anders Arpteg:I'm a lot of good learners there. Okay, so for the philosophical question here, imagine AGI is coming. Either, I think OpenAI they released or they leaked this document saying they have a plan that, like in 2027, they will have AGI in place. Ray Kurzweil says 2029. Other people think in 10 years, 50 years or never. Anyway, imagine AGI is here.
Anders Arpteg:What do you think the world will look like? It could be either in two extremes. One extreme is, of course, the dystopian kind of nightmare that we have in Terminator, matrix, et cetera, where the machines are trained to kill the units, and the other extreme is the utopian future where we have the world of abundance. We have no need for energy, energy is free, we have fusion, energy fixed. Everything we have, you know, the medicine is awesome, we have cure for cancer, we have educational systems that are, you know, amazing. We have no need to work if we don't want to, we don't have to work 40 hours a week anymore, and so forth, and we're free to pursue our passions and creativity in different ways. You know, that's another extreme. Where do you think on a spectrum we will end up when AGI will be here? If we start with Matias, perhaps?
Mattias Altin:I hope for the latter, but given all the terrible things happening and expanding in the world right now, with the war in Ukraine that is, I'm very for. The first. I read the.
Henrik Göthberg:But do you think we should be scared of the extremes or do you think there is going to be somewhere in the middle?
Mattias Altin:We don't know right.
Henrik Göthberg:No, it's completely.
Mattias Altin:I read the Live 3.0 this summer.
Henrik Göthberg:Yeah, I read it too this summer.
Mattias Altin:And that's kind of scary.
Henrik Göthberg:But it's interesting how it does the different scenarios right. Yes, exactly.
Mattias Altin:And those scenarios in the book is kind of, yeah, they're plausible.
Henrik Göthberg:But we're talking about Max Teg Max's book Live 3.0. And now, for anyone who read it, I'm trying to recollect. Were there any of these scenarios that you thought was more plausible?
Mattias Altin:The whole thing about being able to trade. I used to work in the finance industry and, like the other, go trading is kind of never, but maybe something smart enough will come in and really tip the scale. I don't know, I don't know, I have no idea, and the scary thing will be we wouldn't even realize it until it's already there. Right, it will be very stealth until one day we realize that that wasn't actually a real company. It was just an AI with some people.
Henrik Göthberg:Yeah, so one of the scenarios that Max Teg must have. He's like in the introduction of the book. He's doing like a little short story, which is like how an AGI takes over the world and the core topic is a little bit like it's stealth, Like people don't know that. So the AI is doing it in such a way. So people are thinking they're working for companies or they're dealing with companies, but behind it all is a shell company and shell company that ultimately they're it's a science fiction. But okay, what about you, Nicholas?
Anders Arpteg:How close are you closer to the utopian or dystopian future if you were to just put some kind of yeah, looking at the history of mankind, sadly I think it's a risk and we have to deal with that.
Mattias Altin:I think Henriquesson thinks that, how we could make it to be do that, but we need to be mindful about it. Yeah, let's go.
Anders Arpteg:Nicholas.
Niklas Modig:Yeah, I mean I hope that the good will win. But I mean it's like my friend. He built a pool and this summer his daughter fell in the pool and it took him one minute or so to see her and he jumped down and brought her up and she was okay. But he said it's the worst and the best moment in my life, because I felt the feeling of losing her and all of a sudden, all the problems that I made up weren't any problems and I wish that the world can get into a situation like that, where our daughter falls into the pool and we think that she will die, but she doesn't. With AI, so we can see what it could do, but it forces us to become the good dad. So I really hope that that will happen.
Niklas Modig:But we need the daughter to fall in the pool, and that hasn't happened yet, what I can understand, but I hope that it doesn't take 20 daughters until that happens.
Mattias Altin:No, therapeutic level no, no.
Henrik Göthberg:And what about you, Henrik?
Henrik Kniberg:I kind of choose the path of optimism because it's too depressing to not. I put it this way we didn't eradicate humanity through nuclear weapons. We've had them around for 80 years. We should be gone by now, but we haven't. We managed to not blow ourselves up, which is quite amazing. We had some close calls. I choose to assume that we're not going to blow ourselves up with nuclear weapons. If they do, then I probably won't be around to miss myself, so by just assuming things are going to go well. If they don't go well, then whatever. That's just too bad. But I assume that things are going to work out for the good and I want to do as much as I can to help that happen.
Anders Arpteg:Choose the path of being an optimist, yeah, but also realizing the risks, I guess, is what it boils down to.
Henrik Göthberg:I like the combination of choosing the optimisms but not being naive.
Henrik Kniberg:Not being naive. Yeah, assume the best and then do everything you can to make that happen.
Henrik Göthberg:What did they say? We hope for the worst, but we plan for the best, plan for the worst kind of thing.
Henrik Kniberg:But in practice, what I think is going to happen is this technology can give us utopia and it can give us dystopia. In practice, the pragmatic side, I think we're going to get a bit of both.
Anders Arpteg:Yeah, we're going to get a bit of both. I like what Sam Oldman is saying about this. This changes view of a thing in the last six months or something, but what he's basically saying. Of course it will be a huge impact on society, but it will not be as big or as profound as some people think. It will be some kind of slow process, not a fast takeoff, potentially, but it won't perhaps be as extreme as people think. I hope he's right. But who?
Henrik Göthberg:knows I mean there's this core question that Max Tegmach explores in the book, which is quite interesting. When we get to the point of AGI, is that going to be a very slow evolutionary path or is there a real singularity where something really takes off in minutes and hours and I think the most pragmatic understanding of it is that it's reduced speculation? But I tend more to like oh, I guess we passed AGI two years ago. I'm not sure it's going to be.
Henrik Kniberg:Yeah, same thing here that.
Henrik Göthberg:GPT was a huge stepping stone right. Was that AGI. No, we're going to have several like that on the way.
Henrik Kniberg:I think it's a continuum.
Henrik Göthberg:It's a continuum, but it's going to go faster and faster, so it's just going to be a blur right, and I said many times before.
Anders Arpteg:I'm more afraid about the stupidity that we have today. When we actually do have proper AGI, I will be very feel safe. But I'm really scared about the time until we get there.
Henrik Göthberg:The fuckups on the way, yeah.
Mattias Altin:A lot of false summits in Montenegro. You talk about when you see the summit and you come up with actually the next one and next one.
Henrik Göthberg:Yeah.
Mattias Altin:We're probably going to be like that.
Anders Arpteg:Awesome, so many interesting discussion. I hope you can stay on for some after work, where we have even more interesting discussions and speak about stuff off camera. That will be, I think, very, very interesting. So thank you very much. Matias Altyl, Niklas Mody Modik and Henrik Nidberg Thank you.
Henrik Göthberg:Thank you, thank you, thank you guys.