My Six Month Review and Opinion of TR AI - Okay but also not

Training
1

I wanted to give TR my 2c about how this is working for me. There are some things I really like about it, but when it comes down to it, i have no idea what the AI is doing because I cannot make sense of the ‘decisions’ it’s making, because it’s not really making me faster. I have had these concerns for months, but I wanted to get through my race and really give TR a chance because training takes time, and maybe I’d see it improve as i went through a couple of blocks.

TrainerRoad Here’s a link to my training calendar/history

Brief background. I started training hard at the end of 2021 to do a 100 mi ultra-xc mtb race. I never got my goal, but i did lots of 50-66 mi races where I placed very well (~6-9h events). I had a coach, I did 8-12h of training a week (5d) and in 2023 added more weight training to the plan before it all fell apart with him overtraining the fuck out of me. My talents as a rider is that I have an inexhaustible aerobic engine. I can hold 85-90% of my FTP for an ungodly amount of time, with my training on weekends often consisting of me holding tempo 80-90% for 2-4h in 30 min blocks with 1-2 mins of rest in between, with these not feeling overly difficult. While the low end of my power curve does drop its more of a flat line than a reduction once you hit about the 2-3h mark: my 2h power record is 260 np, my 5 hour is 242, and my 7 hour is 210, all done with an FTP around ~290-300 in 2022. On the flip side, my 3, 5, and 10 minute powers are absolute garbage, with most of my friends easily doing 30-50 W more than me. I never really seem to not be able to complete workouts, no matter how tired I get, and i also perform my best powers under a pretty significant amount of fatigue. This unfortunately is probably why i overtrain so easily, because i can dig myself into a pit i can’t get out of. At the middle of 2023 i burned out hard on racing, and basically lost my love of the sport for a while and didn’t really want to train.

Now that said, while i’m experienced with training, i’m not THAT good of a cyclist, and between getting older, not having the same training roads with a big drop in volume, a new medication, and more work stress, I am not the same cyclist I was a few years ago. The last year I have had big issues with chronic tiredness, and I do not seem to be able to recover and so I do a lot less volume. When AI FTP came out, my goals were to hit a very consistent 3 workouts a week, with a bonus longer endurance mtb ride for fun if I wanted to do it. I also climb 2x a week. I selected a 3x week balanced intensity FTP builder plan that runs for another week or so. I tried to add lifting (squats and mb deadlifts), but I was cooking my body and physically was falling apart and had to drop it after just one 4 week training block – I couldn’t recover and needed to focus on hitting my marks in cycling.

My FTP generally is ~265-270 when I’m semi-fit, but AI FTP started me off around 245 (actually 220 but that was bullshit and was never correct). I have had a couple wiggles, but i have basically been consistent week after week since Jan 1. The only stuff I’ve skipped (generally) have been some of the easy recovery rides on recovery weeks. Because i know my limits, while i have sometimes selected slightly harder workouts, i have stuck with the recommended workout. My primary power meter is a quarq on my XC bike, and my training bike has a left sided 4iiii. I realized recently that the 4iii is probably reading a fair bit lower than the quarq, and because i trust that more and have more right leg power I adjusted the scale factor a few weeks back to be 5% higher on the 4iiii. I think it’s probably still 5-15w low, but the power zones are at least in the right ballpark.

HOWEVER, because i know my limits pretty well, for a huge number of the workouts I have been doing 10-20W higher than than the recommended power even on the 4iiii, for example the high end of the workout today was 276 (4x5 105% FTP with it asking me to do 266 w) , but I was able to do them at 288, 295, 292, and 284 W. If this was on the xc mtb the numbers would have probably be in the 300s. I only have three workouts a week, so those workouts have to be hard enough because my volume is so low that there’s no way i’d get enough training stimulus. You might say, “well the AI is trying to hold you back so you can get better training, and you keep going too hard so you’re too fatigued”, but the problem is I have been able to do this for months, not fall apart and still complete the workouts. I personally have probably seen a pretty improvement in fitness, but I’m also not sure if that’s because of AI workouts, or just because of the fact i’m training consistently and I’m also not convinced that it’s not just my ‘baseline’ level of fitness that I seem to get to and just sit at if i ride consistently.

As far as what TR says, every single AI FTP assessment has stuck me at 250 W, EVEN though I keep doing the workouts above the recommended level, grade their difficulty based on what I do, and have put out an ability that clearly outstrips the fitness the AI wants me to be at. I am clearly impacted by the outdoor rider bug where the estimated FTP gain starts high and eventually narrows lower and lower and by the end of a 28 day block i’m right back around 240-250 W. My workout levels have barely changed in 6 months, and I keep getting fed pretty much the exact same workouts week in week out (I made a post a month ago complaining about that).

To boil it all down, I think AI FTP has me plateaued, and would continue to keep me there, without me smashing my workout numbers ever time i go out for the harder rides it prescribes. I can’t see any progression in its workouts, or if it is adding progression it is super minimal. Maybe it sees my week-in-week-out numbers going up (which they sort of are), so it’s keeping me at a level that it thinks is “optimum” based on Nate’s quote of “The only thing the AI model knows is watts”, but i struggle to see how ITS doing that, because the only reason i’m hitting higher numbers is because i’m not doing what the AI is telling me to do. Even with me trying to force it/tell it that, “i can do more power and more load than what you are giving me!!!” by going out and hitting these numbers, no matter what it sits there and feeds me the same 250 w and approximate training levels every single week.

I can’t entirely complain, because i DO think the workouts its giving me are beneficial in structure, but not in the specified loads, DEFINITELY not in power, and DEFINITELY not in progressive overload. I am reasonably happy with the new system compared to older adaptive training, but frankly overall i am pretty disappointed with the new AI model, because I can’t see what it’s doing that is much better than the program before it. For example, the race i did a month ago i did i think 215 NP for 4 hours. I would have assumed that the new ability of the system to ‘analyze unstructured rides’ would recognize say “oh hey, this guy we think has an ftp of 244 might be a bit stronger than we thought if he can do a 4h mtb ride at 88% of FTP”, but nope, my AI FTP came out the next day and actually dropped my power lmao.

Anyway, I wrote a huge wall, not sure if anyone will read this, but I hope TR can take it onboard and tweak their system to be more helpful and effective. I do really think there’s promise here, but the ‘black-box’ model to me quite clearly does not seem to be adaptive at all, at least to me and my fitness and abilities.

EDIT: one other thing i hate about the new AI FTP is it totally fucks up our TSS. My 20 min power is CLEARLY not as low as 250, but whenever i train it’s going to calculate TSS’s as if i’m a monster. So if you see my training load (as TSS) in my calendar for the last few months you have to take that into account, because everything before that is actually calibrated correctly. E.g. my loads now are lower than they used to be, even if the bars look like they’re high.

2

Thanks for taking the time write your post. My history and experience is similar. As a result I’ve broken away from the TR Blackbox and am just using it as a workout library. All a bit sad really.

3

I’m wondering about these two statements you’ve made.

Having pointed that out I do get your point. I’ve wondered a lot about TR’s fatigue detection algorithm. I think it just doesn’t work for everybody and probably holds some people back too much. I mean we are all individuals and even though this is the All Seeing All Knowing “AI” it is a new system and it is just math. We are not algorithms however and everyone responds differently to stimulus so it remains a mystery to me how much to really defer to the AI.

For myself I’ve just learned that I can push the fatigue harder than TR thinks I can but I keep the workouts exactly as prescribed as far as power goes. So I will pick a longer workout when I feel like it but let TR adjust that workout’s power as needed. I never really touch the +/- % power though unless I’m in a “bail out” condition trying to finish a ball busting workout.

Basically I’ve learned which components of TR AI that I can rely completely on and which I can fudge with a bit and tailor more to my specific needs. I think most riders would benefit from a hybrid approach like this. Tinker with the settings and learn how to make the system really work for your specific needs and biology. It’s all part of the fun of it IMO. It’s an incredibly powerful too. Use it, don’t be used by it.

4

The fatigue I think may have been that I got covid again or something last winter. When I got sick in May of 22 it absolutely wrecked me (“long covid”), even though i didn’t get that sick. I have noticed a distinct improvement over the last several months. Also, I don’t want to go into details at all, but tried an alternative medicine for a chronic issue last year and that did some weird shit to my physiology - frankly it felt like all of my chronic training just evaporated. It may have been due to detraining, but either way my fatigue is ‘different’. That’s why I kept myself to 3x/week, basically half the volume i was capable of doing ~2 years ago, because I know I can do it and recover appropriately.

As far as TR red light green light. The pre-AI ftp i actually think had my fatigue down great. The new red light green light? I generally ignore, the only time it’s right is when I have done absolutely massive rides where it tends gives me two red days. All the yellows, let alone reds, are days I generally can still complete a workout, especially because i have 4 other days in the week to recover. 3x/week is not a heavy load at all.

5

I generally ignore yellow days myself as per my comment earlier about fatigue being somewhat personal to our own biology. I probably won’t bury myself with a 2 hour over under session or a 7 hour ride or anything on a yellow day but I pretty much train normally. Yellow days never come up on those days though anyway because they are always preceded by a rest day in the plan. Red days I respect every time except when I’m on a bike trip. In that case I’m riding right through the red days, nothing I can do about that.

It is weird that you feel as though TR is underprescibing your workouts. For me its bang on. More often than not - it happened this morning! - I go into the hard workouts a little nervous that I won’t complete it, exit the first interval going “oh no dear God how am I going to survive this” and then end up finishing strong. The perfect balance on the knife edge more often than not.

6

I suspect the AI was trained on all user data. To what degree it uses each user’s personal data to make predictions unique to that user is anyone’s guess. But I did get the feeling over the last few years that TR felt it was overtraining everyone so they seem to err on the side of less TSS. But that isn’t right for everyone. Some, especially those with decades of aerobic training, can benefit from tons of volume.

7

100%

Great to make you fast. Once you’re fast the bell curve is going to keep you there, smack bang in the middle. Plateau town.

8

Who was the coach? Always looking for reviews of who are good and bad

9

Agree with so much of what you’ve experienced. Came back to TR after reading and hearing about the new AI system around Christmas time, I’ve since cancelled but rejoined again last week (I was off work and bored on a very wet day), rather than jump straight into a plan I thought this time I’d give the system a nudge and pick a few workouts to show it where I am.

I did Emerald-3 a VO2 max workout,2 sets of 2min intervals with 3min rests and 8min recovery (I think) between the sets, rated it as very hard but I really pushed on the last interval as my heartrate hadn’t really gone much above my LTHR during the intervals up to that point. In your post you mention progressive overload, and this, as far as I understand it, is the bedrock of training, so I was surprised (and disappointed) to see that if I placed an ‘AI’ VO2 max workout on my calendar for a week later it reverted me back to Goat Citadel, 1 min intervals with 2min rests, how is that progressive overload? Also pretty much everytime I’ve built a plan or tried to use the AI workout options, Goat Citadel is thrust to the fore. It very much seems like it is picking from a small group of workouts, I looked for another workout myself, found one with 2.5minute intervals with 2min rests again with a recovery block in the middle, think it was six minutes and I can’t remember the name of the workout, the AI told me that this wasn’t recommended due to the intensity, IT’S VO2 MAX it’s supposed to be intense! How, when I’ve completed a workout with 2minute intervals does it come to the conclusion that regressing me back to 1 minute intervals is progression?

Ultimately, I’m currently building my own plan in another app at the moment, I may continue to use TR for now for the workout library, just to save time building my own, but in it’s current state I just can’t understand how I’m supposed to trust it.

10

A lot to digest here, is it the software, maybe, maybe not. Sounds like there are possible health issues to explore.

Usual caveats, not a coach, etc. and looking at the Calendar, I’m not hugely surprised you are holding fitness and not gaining. Last week, zero rides logged, this week has a 4 day gap and while today isn’t over, that is 2 workouts in 13 days. Quite a few other weeks have 1 or 2 workouts. There are other gaps, like 13 days for travel - it’s fine, life happens but this is at odds with building fitness.

In part, it’s this history that is likely giving you so many Red/Yellow days, it takes weeks/months of being consistent for the system to lower the guard rails.

The recent deep dive is worth a listen on consistency, in particular the concept of the ‘worst week’ and the impact this has.

11

I’ve found that in the grand scheme of the plan, not every workout needs to be Very hard, it is likely taking you down a little to build you back up. This has been my experience with O/U and Vo2 to much success. Hitting those first workouts was a struggle (where the system has come up with a prediction on outside data), it has then taken me down and built me back up higher over 3-6 weeks.

12

This is a direct result of calling something FTP, when TR says it isn’t FTP, but then using that number to calculate existing recognised values that are based on actual FTP. The whole thing is a chaotic mess, either build a system on FTP (as the majority understand it) or build a new system and use your own proprierty values such as the likes of Xert have done, but putting a new system on top of an existing one is just causing confusion.

13

OP’s point about AI decision making is really important. I’d say it rises to the level of a bug.

TR makes changes to upcoming workouts and to your AIFTP prediction and there’s no guidance on why. That seems like a pretty clear “not working as intended”.

14

All of these are outdoor workouts, I wonder how much the AI outdoor bug are affecting these.

15

Oh wow. Y’know, this is on me but I didn’t actually look at the OP’s calendar. Well THERE’S the reason for all the red days and the plateau. That training history as it stands in recent months is not at all conducive to forward progress. It looks exactly as I would expect it to look given his complaints about his results. Many days off then burying himself for a week. Many days off then burying himself for a week. Repeat. No real mystery there.

16

Appreciate you putting your hand up, I often open these threads, read the comments, look at the calendar, doubt myself and close the thread :sweat_smile:

17

Usual caveats, not a coach, etc. and looking at the Calendar, I’m not hugely surprised you are holding fitness and not gaining. Last week, zero rides logged, this week has a 4 day gap and while today isn’t over, that is 2 workouts in 13 days. Quite a few other weeks have 1 or 2 workouts. There are other gaps, like 13 days for travel - it’s fine, life happens but this is at odds with building fitness.

Oh wow. Y’know, this is on me but I didn’t actually look at the OP’s calendar. Well THERE’S the reason for all the red days and the plateau. That training history as it stands in recent months is not at all conducive to forward progress. It looks exactly as I would expect it to look given his complaints about his results. Many days off then burying himself for a week. Many days off then burying himself for a week. Repeat. No real mystery there.

I want to make it absolutely clear. I am not burying myself on the weeks i am given workouts. I am going far above the power targets, AND absorbing the training and feeling fine. If i wasn’t, i wouldn’t be able to maintain this for 6 months. This is a low volume plan, with me moving the workouts around of course there’s going to be big gaps between the workouts. I put 3 days in out of 7, I am not expecting to go semi-pro on this: i wanted to see slow steady progress back to fitness, something I know i could easily hit the marks on and be consistent.

As far as the other comment about missed workouts: the majority of missed workouts are recovery weeks, where quite frankly with the heat in DC i just don’t give a shit about riding around for 50 min in z2. If im going to ride or do something in a recovery week i’ll be my own choice, and i’ve often chosen easier climbing or something rather than TRs workouts. I have often added my own stuff so that my rest weeks were more like five days not 7.

The final thing that I want to make clear to the thread. My problems with TR are not necessarily with my fitness. its how the AI is making decisions about the data that it sees from me. In theory, based on everything that Nate has said, the AI sees nothing but watts and tries to make you faster with that in mind. The problem is the AI has made NO changes to my plan that I would consider within the mark of doing that. It’s feeding me the same workouts, workout difficulty and same volume week after week, outputting the same AIFTP and workout levels.

With my life stress, workout vol/load, yes i may not be able to improve and COULD be plateaued because of that. That’s perfectly valid and I am okay with that. But the issue is that the AI is doing nothing to try to build a plan that could see small incremental fitness improvement given the limitations, and given the limitations AND that I can do more work than the AI thinks i can do to get faster its an even bigger issue. If I am to plateau, it is because of this, because within the space of 3 workouts a week there is no progression.

Oh wow. Y’know, this is on me but I didn’t actually look at the OP’s calendar. Well THERE’S the reason for all the red days and the plateau. That training history as it stands in recent months is not at all conducive to forward progress. It looks exactly as I would expect it to look given his complaints about his results. Many days off then burying himself for a week. Many days off then burying himself for a week. Repeat. No real mystery there.

I already told you above that you can’t trust the red light green light with my data. Those red days are not often red days based on how I feel unless it’s one of the 3h+ rides, and the yellow are completely wrong. The only times I have truly felt out of energy have been on the rest weeks for 3:1 on/off. Which is how it’s supposed to feel. As i said at the top of this post, its a low volume plan, there’s no “burying” myself on low volume.

18

The AI is saying follow this plan - I’m not sure what else it can tell you, here are two things I would pause and think further about:

  1. They are not recovery weeks in the sense that you do minimal work, they are deload weeks, you can’t skip them if you want to improve. The AI is telling you that you need to do these.
  1. RLGL isn’t a ‘body battery’ or ‘how do you feel today’ measure, it’s a long term predictor of fatigue, worth reading the FAQ https://support.trainerroad.com/hc/en-us/articles/20753792083995-What-is-TrainerRoad-s-Fatigue-Detection because if you do feel it is being too conservative, this can be adjusted in your user settings.
19

They are not recovery weeks in the sense that you do minimal work, they are deload weeks, you can’t skip them if you want to improve. The AI is telling you that you need to do these.

I disagree. If the training plan and fitness improvements hinged on ~3h of z2 every 4 weeks then there are bigger issues with our understanding of improving fitness.

20

You use all the right words: “slow steady progress”, “consistent” but I can’t see anything steady or consistent in your calendar. That’s all I’m trying to communicate to you.

And there is a huge difference between a deload week and a rest week. If you are skipping 50min in Z2 on a consistent basis then you are absolutely negatively affecting your goal of “slow and steady / consistent” progress. I’m sorry but you’re just wrong on this. Keeping your aerobic base strong is foundational to cycling fitness and it takes hours in the saddle. Skip it if you want but don’t expect stellar results.