Raggiungere la tua forma migliore con l'IA

Niclas: Welcome to the Afasteryou Podcast, where it's all about endurance sports and training. Sebastian Schluricke, Björn Kafka, and Niclas Ranker bring you valuable tips and insights to help take your performance to the next level. Good morning, welcome to a new episode of the Afasteryou Podcast. This is the first episode where everyone involved in this podcast — or at least everyone named in the intro — is actually here. Good morning Sebastian, good morning Björn.

Sebastian: Morning. Yeah, very good morning. Nice to be here.

Niclas: I've actually been asked a few times — yeah, who is this Sebastian? What does he even do? Why is he always mentioned in the intro? Sebastian, what do you actually do?

Sebastian: I sit in the basement and work.

Björn: Sebastian is the heart and the brain — and not the people who swim along in his slipstream and bask in the success.

Niclas: Sebastian is the only one in this podcast who actually works.

Sebastian: That's true. I was on the podcast in the early days too. Björn and I started it, and we actually did a few episodes back then. Then Niclas joined, and it quickly became clear — we can replace Sebastian.

Björn: Sebastian is the boss of Afasteryou. I just have a few shares — or a few more, too. Sebastian started with Aerotune. We then renamed it to Afasteryou, shortly — or somewhat later — once we had all the metabolic stuff in there. The whole thing started with aerodynamics. We've definitely explained this in earlier podcasts, but just to bring along some other listeners who joined over the past months and years. Sebastian is above all the person who programs a huge amount. I'm in active exchange with him on simulation models, ideas for how we can do training, and so on. We're constantly in correspondence. Numbers guy, programming guy. That's how it is. Sebastian, feel free to add something.

Sebastian: I think that captures it well. We all have our part in the company, and everyone has their expertise, and all together it makes us strong as a team. And that's what's truly special, and what matters to me — that as a team we try to bring all our strengths into this company so we can offer something positive to our users. I think we've actually built and created something real now to deliver real value to athletes out there. My main share is definitely that I program a lot. Now also with the support of agents that take over a lot of programming work. Some of you may have noticed. We've built significantly more features in the past weeks — pumped them out, almost — than before. There's always a certain trade-off when developing software with AI agents. You're significantly faster, but quality suffers a bit. Still in a way that we can keep shipping good products. When developing with AI agents, you have to make sure they don't work too sloppily. Because an ecosystem like ours is pretty complex. We have a lot of servers, over eleven, that make up our infrastructure. And we have certain standards. You probably all know it from ChatGPT and similar — if you stay in one chat too long, you can get hallucinations. It makes something up that wasn't actually said. Ultimately it's based on statistics — it takes the highest probability of what should presumably apply now. And that's the problem in very large ecosystems like our infrastructure — that it suddenly wants to install or program things in parallel that we don't even need, because we already have a system for it. It just forgot in the course of the context. So you have to handle these AI agents cleverly when developing software. But it has enormous possibilities — that we can offer significantly faster and significantly more to our users. One of the great things we've recently shipped is our Virtual Coach, still in beta, but it has tools at its disposal. Unlike many other large language models that just interact, our Virtual Coach has real access to the database. So... ...the user can get information from their training plan via the Virtual Coach. They can restructure their training plan. They can say, no, I don't have time, push the training to tomorrow. Next week I always have two days on Tuesday to train, can't we build in a double session? It can do that. It can restructure competitions. It can plan secondary races. It can replan the main race. It can mark you sick. It knows all our data structures — basically it has access to our system. We now have over a million analyzed activities. We have pretty good statistical evidence of what really works in training, data-driven. And it understands these concepts. So often the question from the athlete is just — why are certain things in our training the way they are. And it can explain — because you can clearly say, you don't have to take over all the standards from training philosophy that have settled in as best practice. We can clearly say the data shows other things — for example, it makes sense from the start to train and stimulate VO2max. Athletes benefit from that. There are also relationships — how does VO2max change, how does lactate production rate change. The Coach has access to all of that. That's what makes it special. And what I find really cool — and this is the last one, before we keep going on monologue — you can chat with it via WhatsApp now. Which I find really cool. And soon things like — when you've completed a workout, it'll directly ask you over WhatsApp, hey, how was the workout? And you can give direct feedback. Also via voice. You don't have to type, you can send voice messages. That works too. And that's pretty cool. There'll be more on that. But that was a lot of UI talk. There are obviously many other aspects too. But that's just a small excerpt — or large excerpt, depending on perspective — of what I've been doing and what we coordinate together, whether it's actually products or features that offer real value to you.

Björn: Let me jump in real quick. Now everyone wants it. The thing is of course Sebastian and I work closely together. Staying with the Coach — you can ask it really smart questions and you usually get a really smart answer back. Real metabolism-specific questions about training and so on. Because it knows the things we built in, that we taught it. And that's really cool. Sometimes — I had a case recently where I thought, okay, why does this behave this way? Why does glycolytic power jump up now? And then I talked to our Coach about it and thought, yeah right, exactly that I have somewhere deep in my brain too. But I'd forgotten — the machine just knew, because we trained it on it. But Niclas, you...

Niclas: Yeah, well — we've gone really deep into the topic without bringing the listeners along first. So we're talking now about the entire Afasteryou platform. Okay, we started with Aerotune, originally just Aerotest. Björn joined and basically... ...brought in the metabolic profile, with which we then started doing Powertests. Most users probably know the Powertest. Sprint test, 4 minutes, and 12 minutes. And you get your VO2max, your VLamax, your Critical Power, your MLSS, your Fatmax, and you can already start training with that. Then — as far as I know, one of the biggest things we offer especially in pro teams is also a lot of velodrome testing. So aero testing is still done a lot. That's also where Sebastian and I were at Wismar on the velodrome together, because they work a lot with the Aerotest function. But that's used mainly less by what I'd call the normal user and the normal podcast listener — versus the Powertest, for example. It's still a strong tool you can do relatively easily. Yeah, you can just do it outside. Right, you can just do it outside, you don't even need a velodrome. You just need a kilometer of straight road, ideally. So — what other possibilities do we have on the Afasteryou platform now? Because so much has been added recently. Aero tests, most know. Powertests, most know. But about the rest we've talked relatively little so far. Björn keeps mentioning bits here and there about everything we have. But now that we have Sebastian here, he can explain step by step everything you can do with the platform.

Sebastian: Yeah, theoretically I should be able to. Not just theoretically — practically too. You're absolutely right, those were our main products — the Aerotest was our main product back then. Then through the Powertest, the Powertest definitely became the main product for most of our customers and athletes. Still very interesting for pros, especially teams. In parallel for those teams, the simulation is also very interesting, because we have simulation software with which you can do race predictions and try out different pacing strategies. We're now at the point where you can compare four pacing strategies against each other in this simulation. You can pick different metabolic profiles. You can do a normal W-Prime model for pacing strategy. You can also use the Mader model. You can input different aerodynamic coefficients. You can also implement lean angle for corners and braking decelerations. We're well set up there, and the WorldTour teams use it to plan their time trials and develop optimal strategies. But that's again very specific, not used much by the broader athlete base.

Niclas: Right — that's what Björn does during the Tour, for example, to predict a TT time. He hit it almost exactly at the last Tour. I think Björn was within a few seconds of Pogačar's time, or Remco's, perfectly. Or for example to do an Ötztaler simulation. Okay, what's theoretically possible at the Ötzi? If you have the FIT file from the Ötztaler and a current Powertest, you can play around with this simulation and look — okay, how fast would I be roughly, how long do I need? Maybe from that you can draw conclusions on how many grams of carbs you should bring, because you're not on the road for 8 hours but 9 or 10. You could use it for that.

Sebastian: Right. And it goes a step further — this tool doesn't just say how fast you are, but also tells you, if you input how many carbs you can take in and your CdA values: ...the system tells you, you have to invest this much power here, this much there, this much here, to get the fastest split. And you also see in this simulation how many seconds ahead of constant power you are, or ahead of another system. Plan metabolism — and you have a way to weigh up what could actually be how fast. WorldTour teams especially use this for time trials, to extract every possible second with different pacing strategies. It really works best in time trials. That's where it has its biggest strengths. For drafting races it has limited possibilities, because it's very hard to predict how much you're in the group versus alone. Then our main tool, where we invest the most time, is actually our training planning. We've had this training planning for over a year now. We let it run in beta for a long time to really see that everything works, took a lot of time to gain experience with this training planning. And today, like many competitors or platforms talk about generally, there's a lot of talk about AI Coaches and AI training. And that's the kind of thing that — I think — leads to a lot of misunderstanding about what AI is actually technologically capable of today. We can certainly talk more about that today, because for us it's always our standard that our users also get some education on what's actually technologically possible and how this is implemented. And how we do it, and how others do it. There are various studies in the marketing space too — the term AI right now is being extremely milked, used everywhere. Theoretically, no matter what you do, you can always say it was AI as soon as ChatGPT wrote one line. AI was involved. But that doesn't necessarily mean those are the best training plans. AI and creating training plans — there's already... ...know-how in there, because fundamentally an AI has read incredibly many books, and where there's a lot of material — large language models are strong — they can draw good conclusions or at least present good summaries. ...they can draw good conclusions in that area or at least present good summaries. Whether it's a good individual plan is always tough, because it doesn't know your VO2max, doesn't know your lactate production rate, doesn't know how much time you have when, doesn't know how you actually executed the sessions. That's where it gets hard for a large language model. Our approach is different. Relatively early on we developed our Powertest AI, which based on the Powertest can predict a VO2max and VLamax — predict from your normal training data. It's also the basis for our training planning, because we can detect whether something in your VO2max or lactate production rate — your economy — is changing, and react accordingly with our training planning. That also has a lot of Björn's know-how in it, because Björn has been a coach for a very long time. He has lots of experience, knows which workout structures actually work. We have to differentiate between what's actually scientifically measurable — training that works. Training in training zones, how much energy do I burn in training zones. You can investigate that scientifically. We can investigate that in our data. What gets harder is fancy workouts with lots of different stuff in them, where evidence that it actually works is hard. You can certainly argue and try to scientifically explain why such sessions make sense. I'm not denying that they can work. But to determine in a data-driven way that it actually has an effect when done a certain way is incredibly hard. We try to build it from the ground up. We have Björn's know-how, who simply has a lot of experience with the top people, and we developed workout structures. Over 200 workout structures. Our AI is specialized in knowing how much energy an athlete needs to burn per week, in different training zones, so we get an adaptation in VO2max or VLamax. And it replans this constantly depending on what you actually do. Our idea was always to find a kind of navigation system for the athlete. Even if you take a wrong turn, the system still gets you back on the right path — because you may rightly say, no, today I want to ride three hours completely differently because I'm out with my buddies. And just like the navigation system lets you take two right turns because you want to stop at the bakery and then go somewhere else, doesn't lose its nerve and keeps bringing you on the best path — that's how to think about our training AI. So you don't lose the fun in the sport, don't get dictated to too much. But if you want it, you have a really good option with this system. And if you want to take a right turn, that's no problem either with our system.

Niclas: Okay — but we've talked about training planning and so on. So the athlete comes onto our platform, does a Powertest, then connects their Garmin, their Wahoo, can upload every session, and sees for each session through our Activity AI: okay, what was my VO2max today, my VLamax, how many carbs did I burn, how much fat did I burn? What influence did this training session have on me? And then we have this AI Coach that you can chat with, that you can have build a training plan for you, that navigates you — as you said — provided you keep telling it: how was my training today, did I feel good? Did I feel bad? It detects on its own how well you executed the training when the activity gets uploaded. The AI can do all of that, or the LLM, whatever you want to call it.

Sebastian: You have to differentiate a bit. On one side we have neural networks. Real artificial intelligences we trained with TensorFlow. That's a big Python library you can really train AI with. We have our servers and we run it there. Those make this prediction on every normal activity, for example. Then we have machine learning. We use machine learning to define the volume per week for athletes — how much energy you should burn per week to get adaptation. We do that with machine learning. There's no neural network there yet. We're working on it. I can come back to that at the very end.

Björn: And what neural networks are — let's talk about that now.

Sebastian: Yeah, it's a lot. We have to be careful not to lose the thread.

Björn: But these are terms that come up all the time outside of Afasteryou.

Sebastian: Good, then I'll do it. I just want to briefly answer Niclas' question. Our Virtual Coach, that's a large language model — it's just the communication interface to our AI technology. The LLM doesn't plan anything for us. The LLM also doesn't create the training plans. From our view, the LLM is only there to — and we see a huge opportunity here — simplify interaction with the platform. To be a contact point that understands which technologies are behind our system. To make interaction more natural, because... ...more and more options keep coming into this platform, and it's getting hard to navigate it, configure all features the way you'd like. How many sessions do I want? Do two sessions make more sense than three? I could do this. Someone has to advise you along the way — normally that's the coach. You can do it with manuals — we don't think that's good. So we decided to implement the LLM, which knows our strategies, can interact with the platform and configure what you want — but the training planning itself comes from our algorithms. Now we can take a step back — what is artificial intelligence at all? Artificial intelligence is a term from computer science and describes methods in programming, how you can proceed. What you all know is ChatGPT. By now I assume everyone knows it. That's a large language model, and it's a neural network. The really special and crazy thing about it is that theoretically and practically it just predicts tokens — predicts letters that fit a certain context incredibly well. And it's totally fascinating that this works. ...you can predict letters and meaningful things come out that can offer real value. That's neural networks. Neural networks — you can think of it similarly. A psychologist would still describe it differently, because our brain is significantly more complex, we have many more perception capabilities, our brain is built three-dimensionally, and there are completely other capabilities in there. But the basic idea of a neural network is also: neurons that can be activated, similar to our brain. And then can be trained. The most common is supervised learning. Simple example you can always use — I show these neural networks images with cats. And images without cats, and I train it by labeling beforehand and saying, this is a picture, there's a cat on it; here, no cat. So the neural network learns, okay, these are images with cats — there's much more involved. You can't just throw the information into a neural network like that. You have to pre-process the image information quite heavily with various filters and technologies, so that the neural network from this set of information is actually able to decide whether it's a cat or not. And that has a lot of know-how and knowledge in it — how you build such a neural network so it can develop these capabilities. But fundamentally that's supervised learning. And a large language model is also a lot of supervised learning. The right answer is given. And then this AI tries to find the right answer based on the data, by developing its own neuron so it gets exactly there. Just like with the image. It learns from the data to find the predetermined answer. And that's how most AI technologies work — supervised learning. There's also unsupervised learning, often clustering methods, significantly more complicated, because you don't know in advance what's there. You could also say — we use unsupervised learning AI technologies, then many parameters come out that we can't initially do anything with. Because the AI of course finds completely different relationships. It wouldn't find a VO2max but other parameters that maybe data-driven make even more sense, but for us as humans would be incredibly hard to interpret. That's why we always decided not to do it that way. The state of the art we have, and the Mader model, is incredibly good because it's very complex and describes many things — so we always rather try to predict these parameters, because we still as humans have a lot of experience to derive something from. And machine learning is also part of the technology. Artificial intelligence is different. Those are algorithms or specific mathematical methods you can use to make predictions based on statistical data, but there are no neurons. Sometimes people talk about strong and weak AI, but everything belongs to artificial intelligence — and is important in many areas. And I said earlier — Björn said we use machine learning to determine this energy distribution per week. How much energy do you need to train so you get an adaptation? We also use machine learning to analyze your HRV. If you're connected to Garmin, for example, we get your HRV data, your sleep data, your resting heart rate. And from this data — especially the time-course of it — you can say whether your HRV is normal or elevated or too low. There are different ways you can interpret it. We have a scoring system, so you can really do something with the value — because just the millisecond value you get from Garmin alone has little meaning. What matters is how it was over the last week, the last two weeks, how you compare to the long-term trend. And often you use machine learning here. Simpler methods, but with which you can also draw good conclusions from very large amounts of statistical data — large volumes — about how fit a particular athlete currently is. We have these daily values. We call it Readiness — how ready you are. And if this Readiness drops, then for example our AI training planning — which we call AI training planning overall, because real AI is in there, neural networks, machine learning is in there: It would then reduce training, say okay, today we do less, today we do no intervals. We adjust the training. In parallel we have Body Reserve. That's also a statistical value we found, fundamentally based on Mader. It's based on protein turnover — how many proteins you destroyed in training. And the more proteins you destroy in training — slightly counterintuitive — the more your body develops the ability to regenerate proteins again. So it stimulates each other. But at some point you reach a point where the stimulation can't be followed up and your body can't repair anymore. You need that stimulation in a certain way for an adaptation to develop in your body. But on the other hand, it can't get too high either. We have good statistical data on our platform now, because we have so much data, that we know where your Body Reserve has to be for adaptation to actually happen — and from when, for example, more training brings nothing more. You can keep training and you see in the data that more volume would actually still lead to effects on VO2max and you can keep increasing it, but not significantly higher than slightly reduced training. We're talking really about huge volume. The only thing that significantly rises is injury risk. These are things that develop with time, because we have so much information about our athletes that we were able to develop such a complex system as this AI training planning, with various building blocks. And we really keep developing it continuously. Currently we're working on the next generation of Powertest AI, to get a significantly better prediction from every bike ride — because that's the key to success, to actually develop a neural network that can predict training for you. And I think we can generate a breakthrough this year — and presumably, that would be our wish, develop the first neural network that has actually individually learned within a digital twin framework on you, understood over the years how you function, with all our approaches, in a way that wasn't possible before because the data volumes were too large and especially labeling was missing. Remember — I said supervised learning, an AI has to be labeled. If you have no labels, you can't train. What's it supposed to learn? But because we have this Powertest AI and can really say week by week how your VO2max changes, we have the ability to train a real neural network on it, and we believe we can really give the users... well, we're very curious. We don't know ourselves yet what'll come out, but the possibility is gigantic.

Björn: Just briefly — use case, yeah? For me as a coach — when athletes, almost all athletes I work with, and there aren't that many anymore — Body Reserve as a good indicator. We did a big statistical analysis: from when do we have positive adaptations on VO2max, from when should I taper training so I'm ready for a race, whether a one-day race or a stage race. And I just look at this value. If I'm in the red zone, bluntly put, I know I have to seriously be careful. In the yellow zone I have a relatively good trade-off in terms of form development. And anything above is nice to have. And when I go into a race I have to be really in optimized for a one-day race, maybe even in a fresh range if I go into a stage race. And that's for me — who has many athletes who don't have an AI-written training plan but a plan written by me — simply this control: okay, how much can I still push the athlete? Yeah, that's the most important tool currently on the platform side. It's really crazy, but I check it every day.

Niclas: Okay, so for the listener — you sign up, do a Powertest, connect with Garmin. Well, you need that to use it fundamentally.

Sebastian: We have a trial month now — you can come into the platform, you don't have to pay anything yet, you get a trial month, you can connect with Garmin, Wahoo, Hammerhead, and we're working on more connections. Hopefully soon many, many more. Connect, then you have four weeks of training fully in there. And if you connect with Garmin or Wahoo, we pull the activities from history and can use this Powertest AI, already know how much training you need. We can determine your VO2max, VLamax. Then there's a training plan for you. In this training plan, the Powertest is also planned right away — included in the trial subscription, so you can do a high-quality Powertest right away. Which from my view is super for every athlete to have done. As long as you ride four minutes all-out, it's great. I always find that realization — when you see for the first time what your lactate production rate is and somehow you go, wow, I have a 0.8 or 0.9 lactate production rate, I'm completely different than I always thought. Or such an extremely low lactate production rate. What have I been doing with my metabolic system the whole time? I just kept economizing. There are correlations we see in our platform — statistical correlations. For example, athletes who lower their lactate production rate, with very high probability also lower their VO2max. Meaning if I always say I want to lower my lactate production rate, you'll always be lowering your VO2max too. That could explain the phenomenon that athletes who come from short distances, bring a big engine, a big VO2max, and then only train on their economy — they lose the ability of that big engine. They make it smaller. Because they constantly only want to become more efficient. So that's the kind of information that comes out. But in this training planning you have the Powertest right away, and I think it's super important. Because it still very clearly determines, per Mader, what you're capable of, and is also important for the athletes on our platform because it gives us the chance to give a real digital twin. Even if we have an AI that can predict based on heart rate values and power values — thought of a bit more abstractly, because the AI also makes it more abstract — we still have the ability to say, okay, that's what the AI does, but this is what the Powertest produces. There are usually deviations. With some athletes bigger, with some smaller. But that's the anchor for the AI, to say okay, here the digital twin starts. This is what's actually produced, this is what I'm predicting, and now I can better understand how the athlete works. That's why it's important the Powertest gets planned in. Then you have basically little risk in terms of financial concerns. And every athlete should be aware: we spend so much time on sport. It's so important to also pay attention to quality. Saving in the wrong places usually doesn't help. Same as with — bike fitting I can do myself. It's always super sensible to go to a bike fitter. Most come back with an 'aha' effect and know — okay, riding can feel different. Sitting can feel different. It can all feel much easier. Same with training planning. You shouldn't try to save in the wrong places when investing so much time in such a great sport. Performance diagnostics, a Powertest, is great. We offer the easy entry. On the other hand — there's so much know-how, so much technology in there. The actual investment is your own life-time. Saying okay, I'll try this with such training planning, I'll engage with it. That matters to me too. I do the sport too. One wants more success, is a competition athlete. Another maybe wants to develop more fitness. A lot of life time goes into it, that's why quality is so important.

Niclas: Okay, but the whole thing only costs in quotes 25 euros, right? Right. So a test compared to a real lab performance diagnostic — I'd say on average a tenth. And so anyone who already rides a bike, definitely has a power meter and HR strap, can implement it really well. From there it goes — you have to tell the AI once what your goals are. Which races do I want to do, when is it all, and then keep regulating in terms of time planning, and give feedback on how I'm feeling. That's basically the athlete's job, and the rest gets handled by the AI.

Sebastian: Right, exactly. The big challenge is once this user input — but it's important. Remember, it's just important for you. We can't guess how much time you have on Tuesday. We just don't know. You could put out standard training plans, but we don't even want that. We want a highly individualized training plan. We want to create highly individualized possibilities for you, because it's just way too much time to do otherwise. Take the time for it. That matters. And then to say, on Tuesday I have the option to also join a yoga group. We now also support yoga, you can plan in yoga, also core workouts. On Wednesday I always have a fixed event, that's my Zwift in there or whatever. You can plan that in. Remember — the AI is like a GPS, like a navigation system, navigates you always to the goal, even if you for example... ...maybe wouldn't be doing the best thing scientifically right now, but it's still your life and you might feel like it. What's also important is to say, today I want to do my group race, and even if it's not the best training ever, that's what it's for, to enter that, and we want to enable that. Exactly that's what we want to enable, this high individualization. That's why we provided the Virtual Coach as an interface, so you can actually implement your complex ideas of how you want to train. And in a very simple, native way — by giving voice messages to this Virtual Coach, and it builds your training plan the way you want it. Our algorithms in the background then make sure that the training stimulus is good enough that you actually adapt and get better. That you spend enough time in the right training zones with workout structures Björn has built up over the years, where lots of know-how is in there, where you can also say — you're not the only ones who trained based on them. You can assume one or another world champion has trained on them too. So there's really something in there. And what you shouldn't assume with AI training is that... I doubt it's possible with the neural network, that it would tell you for example, at the start you have to ride a ramp, then you ride three minutes like this, two minutes like that, seven minutes like this, eight minutes like that — because that would mean someone measured that, that there's some statistic in the background that determined exactly that. And that you can say okay, I now have so much data on it, also labeled, where I can say the effect of this training session is such-and-such, that a neural network actually arrives at the decision based on a huge data volume that exactly this training would be exactly the right thing. From my view that's state of the art — if you look around — de facto not possible. And if someone actually can do it, then chapeau, I just don't see it.

Niclas: And we get to the point Björn and I always have — if a training plan looks too fancy, it can't be right.

Sebastian: Yeah, and there are no studies. Look in the studies. Who has investigated such workouts? Where's a robust study showing such a training has an effect?

Björn: And on hundreds of people, not just on 20 athletes.

Sebastian: I'm not denying that the considerations behind it are sensible. That's how sports science works. The athletes — and especially all the coaches who did the work, who thought it through, who made the observation, who then determined, oh, I have to adapt my training. I can do something with this metric. When my athletes do this and that, they get more successful. I'm not denying that. But to say you can train a neural network on it without having labeled data — I just don't see it. And that's still the advantage of coaches versus neural networks and AI — that they can certainly still make observations in detail with individual athletes that can lead to better training plans currently. Definitely. But it's just a question of time, when data-driven models can be specified and individualized. That's what I said at the start. When we have our neural network for training, it'll be hard from my view for a coach to still present better training plans in terms of volume and intensity than the AI does. Because it'll have actually understood how the training flow or energy turnover in different training zones has to be for an athlete to adapt. Then it'll certainly come down to other factors that the coach can still bring in. But there — I assume we, all training platforms are interested in that, some more some less. At the end of the day everyone will want to do this, because from my view — though we don't know it yet exactly — it'll presumably lead to significantly better quality in training and adaptation. But currently I'd still say a coach with lots of experience can still look at the individual athlete from a different perspective and therefore certainly still present better training in some way. But like I said, it's just a question of time until this volume of how much time I have to spend in different training zones per week, day to day, for the best possible adaptation, that we can solve better in the future with data-driven systems like neural networks. And that doesn't have to be bad. It doesn't have to be bad for coaches. On the contrary, can be very good. I personally even think the role of a coach will perspectively change — that maybe the coach won't anymore... ...maybe not have so many athletes globally, but maybe have more local athletes, maybe spend more time on coaching with the athlete. I think people still prefer to talk to people, not to large language models, even if those have their place. But a coach will, I imagine, spend much more time training in groups, watching athletes, focusing on technique, taking on coaching with the athlete themselves, and certainly still having the overview of training and maybe still understanding one or another detail that such a data-driven system doesn't see, that works on a personal-emotional basis. And then can still say, okay, here's the training plan, I can adapt it. I still know one or another hint the data-driven systems haven't understood yet. And then make such an adaptation.

Niclas: I think it'll just be a big advantage for all coaches who engage with it, to use both together. And I think in the end that's also the advantage for every athlete — that if you work with a coach or only with an AI, you maybe have capacity for other things you didn't have before because you didn't have time or resources. Working on those — you can then say okay, I have my AI Coach, the Afasteryou one, and I can spend the 100, 200 euros I previously spent on a coach per month now on a technique coach and learn to ride better, which works in 1-on-1 coaching, can handle my bike better, maybe spend more time on equipment. It's just a way to get better overall.

Björn: Yeah, I sometimes get asked why I do all this, whether I'm not making myself unemployed. Of course I laugh first. No, I'm not. Everything Sebastian and I do here, it started many, many years ago — I built an Excel spreadsheet where these energies, these flows over years for athlete development were calculated. Far from how fancy we do it now and not with the workout structures. But fundamentally that's what we've implemented here. Energy flow calculations in different zones lead to adaptation X. The coach's role — and at Uno-X for example too — is above all to spend time with your athletes. That's exactly what Sebastian says. You see someone, you see how they act, you see how they react. And you don't even have to see them on the bike. It can just be time at training camp, and then you notice something. We don't train machines, we train humans. So that social component — and you could open a whole other can of worms. How much psychology, how much it influences performance. There are studies that simply show — if you're happy, if you're in a good mood, you're significantly more capable. Regardless of what your metabolic profile says. We also see often athletes who shoot far over the goal where we don't know where it comes from, and you have athletes who from their performance are significantly better but fail in every race, or even fail in a test scenario. There the coach's role is decisive. That's why I see it — like Sebastian says — definitely shifting locally. I assume too. And the AI becomes your sparring partner. What do we have here? How does it look? What's your suggestion for energy planning? Should we go a bit higher here? And so on. That's how I already use it. And that's where the journey will go. Very clearly.

Niclas: I think — once you've worked as a coach with athletes on site — I was at a training camp with Bike Aid athletes, I was at a training camp in Girona with my own athletes. When you see the athlete on site, work with them on site, you can address things, see things — how do they behave during the day, what do they do, how do they train, what are maybe their habits, where you can do fine-tuning. And if you have the capacity for that as a coach because you're also using an AI that already makes the training better — because it can analyze much more data, give you much more input, because you have the complete knowledge perfectly at hand. It's totally normal — for everyone. You're somewhere in your way of working, and I think it's always good for someone else to look at it and say, hey, look at this, you could do this better here. That's my view, and you can discuss it. If you have someone you can do that with — like an AI — that's added value for everyone, and training gets better for everyone. It's just smart.

Sebastian: Definitely. What I always find really important — you can also use a sparring partner with ChatGPT or some other large language model, Claude for example, and use it as a sparring partner. Of course we try to be more than just a large language model with our AI technology — with real data analysis in the background that arrives at decisions. That's also important, because one thing you should never forget — with large language models, that's statistical knowledge memorized. And if you ask 12 times 7, the LLM doesn't calculate — it can't actually calculate — it memorized it and can therefore execute that multiplication. That's why we sometimes see, when odd numbers or decimal numbers are used, that it doesn't work. Technologically there's enormous potential there. If LLMs — they can perfectly or nearly perfectly take over simple programming tasks, also complex programming tasks. They could also just calculate the task with Python. And when that gets more automated, integrated into LLMs — however this ecosystem looks in the future — and they can integrate these tools more easily into their problem-solving, they'll be much more powerful, because they can use software directly to solve specific problems and not rely on their memorized knowledge. And for us — we want that too. The LLM should be the interaction interface with our ecosystem, with our platform, with our AI, and make it easier, more natural in dealing with a platform. Of course I see demand, and we want to finally bring our app to market this year, and I think an app is still important too, also for the athlete. But the future I see more in: it's not the super fancy app where everything is built in. Because for us it's about training, we want to train. Nice graphs and everything is certainly good and has its experience — that makes something special — but basically it's about doing sport, doing good sport, doing healthy sport, getting better, having fun in our sport, fueling correctly, getting our training organized in the right place the way we'd like. And that we can do that in our daily lives. I think the native interaction with a Virtual Coach, the way we offer it, where you can simply interact with voice, will become the more natural way. I think many other websites or platforms in many areas will go this way too — that it's no longer the super fancy app that does it, but LLMs will be this interaction, the communication, the interface to another platform. And what we can say is, the athletes who already use our WhatsApp service with the Virtual Coach are really active. They use it, and they have a totally different experience there, because you can quickly send your voice message. How was it just now? What do I do tomorrow? What does the week of training look like? Can we adjust a bit? It all works. All native. I don't have to search around in an app — where are the buttons I need to press. I can do everything via voice. These are two complex topics. One is this sports-science AI training planning. How do we build really good training plans that work for you? The other is the interaction with this ecosystem. How do I get the information out of it? We have integrations to Garmin, Wahoo, and Hammerhead so workouts get auto-synced and you have them on your watches or bike computers and can train them. But the interaction will increasingly happen via voice messages in the future. And you don't have to use WhatsApp. You can also go directly into our app — it's a Progressive Web App, a kind of light-weight app — but in there it works well too, that you can give voice messages directly and interact directly with the Coach. So you don't have to use WhatsApp, but the option is there.

Niclas: All right. So everyone — try it out as soon as possible, get input to see what you can train better. Connect everything with Afasteryou. Do a Powertest and get going, so in the best case you improve your training. Give the AI Coach feedback to adjust your training daily and ride better, and fuel properly — because it shows you how many carbs you burned, how much fat you burned, what you might need to replenish. I think for many athletes that's a big point — to see okay, today I burned 400, 500 grams of carbs in a big hard session. You first have to replenish that so the next day's training works.

Sebastian: Yeah, exactly. We've made sure to build a training plan that's data-driven on a lot of know-how from the past, that we keep developing, that really makes sure adaptation happens, that you train the right energy amounts, that you don't train too much where the benefit is no longer there, but also train so — have the energy amounts — that you actually adapt. And so everything else can revolve around your life. Multiple sessions per day are possible. You did a podcast on how multiple sessions can be positive in a day. If you split it, also interesting for runners, who want to run morning and evening for example. Build slightly more volume with less recovery overhead. That's possible in the platform too. This training planning offers it. It's a complete system, and we work every day to keep improving it. Every day — week by week we get more features. Hopefully we'll bring cycling and running, duathlon, into the training planning next, so that's also possible — for cyclists who want to run a bit on the side, run sessions are in there; for runners who want to ride a bit, maybe one or two ride sessions in there; and of course real duathlon, which the triathletes can use too. We don't have swimming yet, that'll take a bit longer, but running and cycling as a duathlon combination will be in there. And it'll only take a few weeks, then it's live. Our standard is — we really try to get the best possible out of it as a navigation system for the athlete, so they can map their complex environment in it. Optimally with a Virtual Coach that takes over... ...all this planning within this platform, and then you have something that adapts to your life, and the Virtual Coach that's always there, that you can always message, that always gives answers, no matter when you're out, no matter when you may have concerns that the training isn't perfect right now, or that you've maybe had to adapt — that it can also strengthen you in your becoming. That's our approach. We really try to develop a product. And in our community — Björn and I still look at all training plans, when we manage to. Just to get personal feedback on how everything has developed. And yeah, I'd say get going. The thing is waiting and pushes hard with you.

Niclas: What, Jan?

Björn: No, I just for fun let our Virtual Coach analyze the bike ride I did yesterday. So you don't necessarily need a training plan — you can just have the activities you've automatically uploaded via Garmin, Wahoo, or... ...Hammerhead briefly analyzed. It wrote me what I did and how many carbs I burned, etc. So everything you can see in our little web app. Then jokingly I said, I only ate 1500 calories that day. ...and nothing during training, what should I do now? And the system either answers nothing — depending on how lousy you feel — or maybe ride 45, 60 minutes easy and eat while doing it. Then I asked again how long until glycogen is available again, and there's actually a real suggestion for how many grams of carbs I should eat through the day so the next training in 48 hours, so today with 24 and the next day, until I can train properly again. And it gives a training suggestion too. So that's pretty cool. How many grams should you have eaten? It says — something like 6 grams per kilo bodyweight in carbs would be a pretty good measure. And if I want a real boost, also go up to 10 grams. Nice. It's always fun. Yeah, but we all know it — once run dry or underfueled, then the next day nothing works and training quality drops massively, and the system intervenes really cleverly there.

Niclas: Yeah, talking about that — what does the athlete need to bring for it to really run well? I assume we definitely need a power meter, we need a heart rate strap.

Sebastian: For runners we don't necessarily need it. They need a GPS watch, no power meter — we have our own power calculation in the system for runners. For cyclists, sure. What we do need is always good heart rate measurement, because our AI works through power and heart rate, and those are the two most important parameters. There are others — temperature has meaning, cadence has meaning, everything has meaning, but the two essential parameters are heart rate and power. For runners it would be heart rate and speed. The speed, with grades and gradients, then becomes the power calculation. So a good GPS watch that captures it cleanly. You can use foot pods to use for speed measurement. Various pods you can use to improve measurement accuracy when running. And it definitely helps with cycling — a good power meter, the best ones, dual-sided ideally. I really like Power2Max — those are great power meters. So fundamentally that's the most important. If you'd ask, with what do you have the best experience in our system? As a cyclist: power meter, heart rate strap, then Garmin, Wahoo, or Hammerhead. Getting started with that is the best experience, because you have synchronization to your devices. You have the workouts on there. The workouts come automatically into our system. Our AI can directly evaluate how much energy you burned, how much you should have done. Do I need to replan the next days? Do you get one more interval? One less? Oh, the training was so hard — we put a rest day in tomorrow. It does all that. And you only have that experience with these three devices right now. What we can say is, we're planning to bring in Polar, Coros, and so on. There'll also be an interface to Zwift, we're really looking forward to that. We also want to bring in IC trainer. We have a lot to do here. ...and we want to implement that, so everyone can come into the experience of this automated training platform with the Virtual Coach. And there, if you don't yet want to go the route of a professional coach, you can really already get a top experience and performance increase.

Niclas: Okay, so for example Oura Ring and Whoop are also planned, to get the data?

Sebastian: Yeah, also to get the data. Polar is also really good. Polar has super data when it comes to HRV measurements etc. We want to implement these other systems and gadgets too, to bring the various systems in. What we can say is — we came from this Activity AI / Powertest planning. Most users were Garmin users, that was the first thing we implemented. The second-most were always Wahoo users for us, so Wahoo was next, and then Hammerhead really came up. It was very easy with Hammerhead. Was really good, we brought it in right away when it became possible. And now we see ourselves connecting all the other systems too, because this daily tracking has an effect. Meaning HRV or sleep wasn't good or resting heart rate is elevated — and not just elevated, but our algorithms evaluate it accordingly. Then training gets adjusted right away. So you have a chance that — if you're already dinged — you don't get... ...another nice finishing move with a VO2max session.

Björn: The good thing is the whole package is looked at. It's not just training planning based on HRV, but training planning based on what you did, how you rate the training — that's possible too — and then sleep and HRV resting heart rate. And if everything points slightly down, then sure, training gets adjusted, lightly adjusted. But it doesn't mean — if HRV is once not great, everything gets thrown out. No, it's done like a really good coach would, gently.

Niclas: So in the best case connect everything you have that's connectable at this point and then communicate as well as you can via the web app or normal WhatsApp. Right. Just get going and try it out.

Sebastian: On the phone it's a Progressive Web App. If you go to the website, it asks you if you want to install it. Then it's like a real app installed. It just does data synchronization for what you've done. And it looks like a native app. Not quite as good as a native app — you can't claim that — in terms of haptics and such. But functionally it's all in there. Everything you need. You see your dashboard daily. You see immediately if training has to be reduced or not. You see the session. You see immediately on the dashboard if you have to replenish or not. You see a long-term trend. How have you developed? Is everything on the right track? You have a status message in there. You also see what's planned for the rest of the week. News is in there. There's a brief presentation of the metabolic profile, also displayed directly. You can move over to the timeline, see what you've trained. Strava-like display, with — from our view — more info on carbs, plus rendered curves where you really see where you were, how fast, with time zones in there. Then over to the calendar you can see directly. So there you see closer in calendar view... ...what training is still planned for the week. You can also manually replan, you can also manually say, no, I want to do a Powertest today, you plan it in, gets immediately synced with Garmin, Wahoo, or Hammerhead, you can do it directly. Even if you have no training plan and want to do a Powertest, you can go into the calendar, plan this Powertest directly, then it gets synced, you can just train it. Then you have the Virtual Coach in there so you can chat directly, also via voice messages, and interact with the platform. And you also have all the other options — the Aerotest or simulation, you can configure all of that. It's all in the app. Normal via the browser, via PC, you see more — more comprehensive — but fundamentally everything is there. WhatsApp, the next channel you have. Then we have these two apps as data fields. But that only works with Garmin, right? Only with Garmin. We could do it with Hammerhead. So you see in real time how many carbs you've burned and how well you're metabolizing your fats currently. That's always a bit tricky during installation. Not everyone reads the text. We can't just put out an app that knows how many carbs you burn at a certain power. That's not possible because it depends on your metabolic profile. How would the app know which metabolic profile you have? So in the app, a small access code is always shown. This access code gets sent to our platform. So the platform knows, okay, somewhere there's an app with this access code. And when you go on our website under Connection, you can enter this code, and our platform knows: that's you. Then the metabolic profile from our platform can be sent to this app, and the app knows day-by-day — and after every Powertest you do, or if you set the AI as the basis — what your metabolic profile is and calculates your carbs in real time. But only that way works, because without metabolic profile you can't calculate it. Some are still just installing it directly and don't read, then ask why this strange code is being shown.

Björn: There's no genius solution for that. You have to read.

Sebastian: You can ask the Virtual Coach, it tells you immediately. Yeah, give me the code. It can also unlock it for you. You don't even have to search the website — just say, look, I just installed an app, this is the code shown. Tell that to the Coach, the Virtual Coach, it'll unlock it for you. Anything possible. Anything possible. You just have to talk, then you'll get it.

Niclas: But that's true even with — let me say it — if you get a professional coach like Björn or me, you have to talk to them too. They can only be as good as you communicate with them. So same thing with an AI. You always have to communicate. Right. Yeah. Nice. I think I've answered all questions for now. Björn, do you want to add anything to...

Björn: No, at the end of the day, with such things, if you want to do this, you always have to engage a bit. Invest a bit of time. That matters. You have to give up the partial furnishing of the hippocampus. You have to engage with such things.

Niclas: Nicely put, Björn. I didn't want to say everyone's done.

Björn: No, but it's the expectation that everything is automated. We have AI now, etc. It's not like that. You have to invest five to ten minutes. Same with testing. You have to push yourself physically to the limit so we can find out what you can do.

Niclas: I think — and I actually still see this with many athletes — the AI, the Powertest AI, helps a lot, but a properly executed Powertest is also really important. Optimally try to do it always with the same power meter, with similar conditions, so you can track your progress as well as possible and not introduce many... ...sources of error, because that — as Sebastian said — is the foundation for everything.

Björn: And the other thing — an AI is a statistical model. The experience of what it feels like to ride a Powertest, it can't give you. And that's, I think, the case with very many things. The experience of how something feels, how something should be, the best machine in the world can't explain to you. But I can't take that away from the coach either. No, you have to do it yourself. But a coach has maybe experienced it themselves, if they rode a lot themselves. That's why it sometimes makes sense to have coaches who have also pedaled themselves. They know what they're writing down, usually. So — machines are good, experience you have to make yourself.

Sebastian: I think there's one thing we have to mention. That's what Ali always says, for example. Ali is our marketing expert. She says — people don't always have so much time. And we want to use the time we have for the positive things, like training. Then maybe you don't have so much patience to engage with such training planning. And it doesn't go quite as fast as you imagine. And then you throw it overboard. But I think at this point — this time, you spend so much time on sport. And if it's so important to you, then it's also important that you take a moment to think about the possibilities being given. This short investment you make at the beginning pays off. There'll be more time available for you later. There'll be more satisfaction later, because the training plan matches much better what you have in mind. You'll just be happier. So those 10, 15 minutes you have to invest in such a training plan creation, it pays off, because you'll get the product you wanted much sooner. And then to say, hey, look, I could do two sessions on Friday but I don't feel like clicking around — then maybe later you just tell the Virtual Coach, it builds it in for you. And the more you use the time to engage with this system — and that's general, doesn't have to be ours — no matter what you do, always think about how much time you yourself invest in this area. And whether it's worth taking a few more minutes to think about it, because it might have a high importance in your life. And you say, no, I don't rush this, here I take the time. It's an important topic for me. I spend a ton of time in it anyway. And then I'm maybe long-term even happier with it, because this time I use is just better used. I get fitter, healthier, better fueled, and have more fun with it. And that's the only thing I can recommend. Ali always says — yeah, but people want fast. Yeah, I get it. I get all of that. And it should be fast. I get it. We're working on that. Just remember, if it has meaning in your life, take the time for it.

Niclas: Good things take time.

Björn: Three times not weighing your food and recording carbs and everything in a program — then it's over. Then it's over, right? Anyone who doesn't weigh their food might as well stop. Counter-argument — anyone who starts weighing their food and thinks too much about it, that's the first step into an eating disorder.

Niclas: Hey, you're taking the fun out of it. I like my eating disorder. No, joking. I have nothing else.

Björn: Most likely it is. Anyone who starts actively weighing their food without being a pro athlete doesn't have much else good in life.

Niclas: Oh now — that's steep.

Björn: That was too harsh. Niclas sits in the middle between us. And on my side Sebastian is in the middle. For me it would even be squeezed by the two of us. But Sebastian didn't say anything. He's the teacher here. You're the teacher here.

Sebastian: Yeah, it's like — for those who are in it, it's always good to hear something like this in a certain way. But for those who are caught in something like this, it's very hard to deal with. Everything that touches on disorders or addictions are complex, really complex, and usually tragic things that happen. It's always so hard to come out of it yourself. Absolutely.

Niclas: Whoa, that's a slightly hard end for this episode, I think. Björn, do you have a nice, short, funny anecdote about your son, that he read some bio book or learned something new about fish, to find a softer exit here?

Sebastian: What we can always say is — the AI Coach will always recommend not doing that and always recommend eating diligently, because that's important. And you always have to consider — I've talked to Björn about this — you can of course start, when you're at pro level, weighing things maybe. But the effect isn't as big as you imagine. There are many other areas that bring much more in training than weighing food. It's often about — when you start going into this optimization — because you also want to optimize. And optimizing is fun. At certain points it can get critical, because it's too close to certain addictions. ...or has other causal effects. So you can rather think about optimizing other things. Sport or performance is a huge area. There's so much you can do, and you should rather try to work on the things that have significantly more impact than knowing exactly to the gram how many carbs you took in. Body feel still also decides.

Niclas: I think the important thing with food is also always just awareness. Especially — that's how I always try to teach my athletes — ideally, you build awareness. You weigh roughly once or make yourself aware: okay, what's in this pack of pasta, what's so-and-so many grams of rice. So you can also place these numbers you hear everywhere. Because you have to roughly trace back what's enough food. So they make themselves aware. But it should never end with you only eating when you've weighed it, or only going by it, or only pursuing this one goal. In the end it's always somewhere a balance, and just making the whole thing aware. But it's just as important to sit at the table with your partner in the evening and just normally enjoy the food and just eat.

Björn: As long as you're not fighting for a world championship or Tour de France or something where one kilo really decides, I'd say who cares. As you said, it's good to make it clear once how much energy is in things and that it also makes sense to eat more. We have this nice trend right now that more is being eaten. It's like brushing teeth. Once you've learned, you know how it goes. For example — here comes the anecdote you've been waiting for. I have an electric toothbrush with an AI Coach, sounds super fancy, works exactly one month, then you've got it. But it always shows you exactly where you are with the toothbrush and how clean your teeth are. That's of course super funny, especially for my sons, who you have to force to brush teeth — though slowly they're getting out of that age, they understand it's important. And you get virtually shown how all the plaque gets blasted away and how clean they are, and they shine, and you get points — there's gamification in it. Cool, I want one too. Yeah, it's super. But now you've understood — how long do I have to brush at one spot? Because real proper brushing isn't just shotgun-style across everything. You really say, okay, now just the four molars, zack zack, front, back, bottom, bottom, and half a minute later you do everything in front, and you work through so you've brushed for two and a half minutes. Done. So the learning curve is there. I don't switch on the AI Coach because it doesn't interest me anymore, but I learned it once and I know how it goes, and the result is also feelable with the tongue. As they say — now I'm doing advertising for the electric toothbrush, where have we ended up?

Niclas: Just name the brand directly. The model — people want to know.

Björn: No, we can't, we're not allowed.

Niclas: Of course we're allowed, we're not public broadcasting, we can name everything.

Björn: It's an Oral B, but I think all the others can do it too. It would surprise me if they couldn't. Anyway — if you know how much, so 100 grams of pasta, you know how many carbs are in there. I have to say it again, the Data Field for Garmin with carbs is for me always a bit eye-opening. Then I rode yesterday, power meter was acting up again, calibrated three times, then it worked. Light height is probably also from the battery currently in it. And then at the end of the day there were 90 grams of carbs, was an easy ride. And I ate, how many carbs did I eat? Zero. Yeah, because I thought, easy ride. Now I had to inhale a few Haribo at home, because — oh no.

Niclas: Yeah, that was great.

Björn: Poor you. No, can't be. That was just right. And then I sat there and thought, that's cool now, next time I'll do it a bit easier. I felt a bit flat afterwards. That's not so sensible, because you want to — unless it was a killer training — still shape the rest of the day. Especially if you have family. So the learning is — once learned, never forgotten — that's important. But if it gets excessive — with food and weighing and 5 grams here and 5 grams there — then we have a problem. Yeah.

Niclas: Good. We've got it. We found a nice closer.

Björn: Yeah. Toothbrushing and AI.

Niclas: Perfect. Thanks for listening. If there are any questions, feel free to give feedback. Spotify comments, YouTube comments. To Afasteryou, if you have feedback on the AI, just write directly. And then I wish you a nice week.

Sebastian: Thanks, same to you.

Niclas: Ciao, ciao. Bye.