The problem of using ML is “guessing” positive, neutral, and negative results are based on the training data you have. These results are significantly affected by outside influences… sleep diet, possibly specific foods and stress…
Very generally we can say ride more, get faster. The goal with dynamic training plans is try to beat a static training plan, and a plan itself is mostly the icing on the cake. Once you have your volume, as you get later in your season pushing your self harder can make a noticeable difference. Generally hard training gives you that quick short term boost especially when you are lacking the volume.
So, I would say a plan with various intensities/types, one long ride a week, assuming you are riding a decent amount and not burning you out, with some hard work towards the end of the plan for a few months would be highly optimized, the details of the specific workouts would have a much smaller effect, lets say the last 5% of performance…
So trying to beat an optimized generic plan with something dynamic without having some of the most important data regarding what worked in the past for that athlete or a group of athletes will be a complete guess at best. You are trying to optimize and addition 2-3% or something and sleep, diet and stress could have a 30% impact or more.
So the biggest benefit of ML/AI training is not to optimize performance, it’s really just a guess and likely has minimal impact. Sad but true… So the biggest goal would be try try to match someones expectations and give them reasonable training while still matching expectations. The user probably needs guidance as to why they are getting a specific training and why another training is not such a good idea… Keeping a cyclist following the plan and inspired in most cases the best you can do… managing load so they don’t kill themselves… and try to inspire them to gradually do more training if they want to get faster.
So I would say a coach looking at ML data and talking to an athlete and trying to come up with ideas to optimize could be great. Eventually an LLM (or AGI LLM) looking at ML and asking the athlete questions could produce much better results… all in all athletes either need to be self coached or have a coach to make smart decisions on on to proceed at this time.
I present all this but I am always looking for ideas to get me thinking outside the box… I do believe the right person/coach asking a fine tuned LLM to create 1,000 - 5,000 different rules could be very beneficial.
I should say training load management is important as well the problem is when you try to manage it primarily with intensity this is wrong, it should mostly be managed with duration and work/rest days/weeks/seasons and more minimally with intensity. With lower intensity plans you have more flexibility in load management so it does not need to be so dynamically managed. When you are right on the edge it becomes critically important.
This is a great thinking! But I think, Strava tried AI feedback after each ride and that seem to be a disaster, so it not be easy to implement right…
Remember were I started this thread - how do I inject my Zwift free rides and/or races into the predefined, structured plan? Not easy, but this is a necessity. Problem - same route can be ridden differently, you can do AdZ in Z2 or you can push. Or you can do HIIT on flat route. Again - training like this may work for some purposes and be bad for others. I’d imagine that training for specific event (like race or tour) may need more focus on the structure. And training for fitness and fun can easily take good, long Zwift ride with periods of over/under TP in climbs and Z1 rest in between for longer than regular workouts suggest. Or just do an appropriate pacer ride for Z2. Sounds great, but how do I know if I pushed myself enough to be in compliance with my plan? Did I ride in Z2 instead of threshold block scheduled? How do I compensate (and should I compensate)? So many questions I do not have answers for…
Oh no I realize none of this is easy and Garmin gives feedback that sucks too I don’t want to be another crappy feedback provider… Now some people can get value from any of these and if it inspires them to stay more consistent then “it’s working.” Consistency is rule #1…
The problem is again with expectations which everyone has their own set of expectations… Generally most people want to make training too complex when it is rather simple (kind of…) get out there and ride you bike as much as you can and sleep as much as you can… eat some decent food… Get decent recovery over all. Manage your intensity and volume to avoid over training (or significant over reaching…) Get some variety and a bit of progressive over load…
Remember a training plan does not have insights into your internal physiology… so adhering to a plan is not the ultimate goal. The ultimate goal is learning to listen to your body and optimize… Coach Jack just provides a frame work for logical progression in all these areas (volume, variety, intensity…) but optimizing for you specifically is not done via forcing adherence to a plan that only your brain and body know what is going on internally with you and even verbally you likely can’t communicate (at least with a machine) until you are broken or destroyed or you don’t like it (too easy or does not fit expectations)…
So your expectation might be “compliance to the plan” and matching your expectation might be very valuable to you and help improve the quality of what you get from that plan… It’s just more about the expectation than it is the requirement than you follow this specific plan. Any no one can dynamically give you the perfect workout today that is exactly what you need other than maybe a great coach that could give you towards the better end. If you generally follow a decent plan even if you not perfectly compliant with the exact workouts each week you are going to get 90% of the value from it and if you are going more by feel and not doing too much you might get more value… but we all need a bit of inspiration to force us out of our comfort zone and getting a good level of variety as compared to just going medium hard every training session.
I am over simplifying this but trying to paint a picture if it makes sense… I am trying to show that dynamic training is not about perfect training but about matching expectations and giving reasonable training because without a human asking the right questions each day I don’t think it’s possible to get highly optimized training that makes sense on a physiology level and if a machine goes into the depth required you will just get bored and hang up the phone on the machine… We have more tolerance to go into more depth with humans…
So I believe our CJ plans are good, I don’t think they can squeeze out that last 5% of performance without a wise human making good choices and dynamic training is unlikely going to squeeze much more of that 5% out either even if you are not perfectly adhearing to the plan based on workouts. Now volume/load optimization or red light / green light is an optimization that could help more than fine tuning workouts each day.
I am sharing my thinking, but I am also willing to change it if I am wrong… I have been thinking about this for years and it seems to be getting clearer at the moment.
Yes, and you will never be able to please everyone, unfortunately.
I always wondered if it is possible to get a detailed description of person’s physiology and feed it into the plan management engine… And continue feeding it while training? That would be ideal, but would probably require 30 minutes survey and some blood tests
At the end of the day, I agree that training advice becomes a task of managing one’s expectations, not just providing perfect workouts daily. There are infinite variants of training which can get you to the goal, but the best would be the one you can use and enjoy.
And that’s what I am trying to advocate for - based on measurements and feedback - provide some additional guidance and adjustments.
BTW, when things clear up in your head on someting you been thinking about - write it down. It helps a lot
Yes even with this idea of red/light green light that is partially based on training load, or significantly influenced by. The thing is TSS is not perfect PMC is not perfect and PMC almost everyone uses 42/7 defaults… but those are very generic and volital. Even blood won’t tell you recover status. So you need something to dynamically adjust those 42/7 values… based on how fast you adapt to training and how long it takes you to recover from a training… So it’s all very broad brush… but it sounds like the way Xert has don’t this is a good idea… Meaning let the user correct the model baed on what your freshness should be versus how you feel. I like this idea… When following a Coach Jack plan their is usually only a couple cases when you even need this. We have safety built into the plan… but still overall it’s something we can and should do…
In my opinion, a training plan should always start easy. I train and coach competitive cyclists (at the top amateur level), ultra cyclists, tour riders, gravel riders, and mountain bikers. It doesn’t matter how good they are. They all start with a couple of easy weeks.
After a block of periodization, they start with an easy week again. The easy week(s) is/are an essential part of a well-balanced training schedule that will make you successful throughout the season. I mean the whole season.
Increasing the training load day after day, week after week will burn you very quickly.
Although I just rode a 200km mtb race, and I’m now training for the same race in the summer, I just had a month of relative rest. Just freewheeling. Now that I’ve picked up my schedule, I ride slowly for an hour here and there. I do some intervals, but they are limited in number and duration.
Over the coming months, I will build them to heavy loads again and taper before the race.
A good training schedule is going up and down like a rollercoaster.
Longer I think about, more I realize that this should be the case. See, in my case I never seen this advice when I tried Xert after self-training for a long while - Xert ingested my history, looked at my hours and load and decided to hit me with even more… Could it advise me to get some rest before jumping on a new schedule? It probably should have.
And here is another opportunity for TrainerDay - when people create training plans for themselves, why not to give them this sound advice right away?
You are absolutely right on periodization, I practiced that while riding on my own. It must be highlighted - switching between plans should be done when one period ends (either naturally or just dropped - doesn’t matter), following some rest before starting new plan.
Because 95% of them don´t want to hear that. So that drives them away from the product and is bad for revenue…
And I don´t mean that this is the case here on TD, but it is on most platforms.
Once users start to realize what exactly is going on, they become more aware. It´s a long and hard battle, but we slowly get more feet on the ground.
We do this. Not the advice but the suggestion plan always start easy.
But we let them change it. @MedTechCD is 100% right that I am sure this turns many away. We need to bridge that gap from the opposite side of the other side of the problem.
Yes, the small army that starts figuring it out can appreciate what we offer, but it is the minority and mostly more experienced, which frequently means older. Figuring out to either convince the rest or adapt the plan to meet expectations.
Yeah, that is sad truth… From what I have seen, TD is much better in this, but would still like to see a more prominent explanation of “start easy” while creating the plan - currently it is buried a bit too deep
Well, that popup is an advice, but I wonder how many people get to it
I don’t want to be another crappy feedback provider… Now some people can get value from any of these and if it inspires them to stay more consistent then “it’s working.” Consistency is rule 
