Thank-you for the very detailed response. To answer the question: are my workouts appropriately difficut? So, far I can do them, and not fail, BUT most of them have not been in the intended zone. For instance, my SS workouts are threshold according to my RPE, and the unders, for my O/U, are not under.
I’m cool with trusting the ML, but has TR decoupled the intent of the workout, say SS, and just want you to do the watts and not care what zone I’m in, even if that means I’m doing a threshold workout? In other words, does it matter that the intended zone or energy system does match the workout description?
This is a really good question. I’m assuming here, but TR did say that the AI uses the plan framework to pick workouts within energy systems, so I’d assume SST should feel like SST, and so on, BUT I also know that if you jack your FTP up manually, the system will turn your threshold workouts into SST because it’s looking for a wattage target. In that case, your SST would indeed feel like threshold.
For Bones and me (and others), we started with a too high detections, and that’s the basis of his question.
For myself, I decided to turn down my FTP manually, and I’ve been extremely happy with the workouts. The AI is serving up great workouts and I’ve already made gains in my SS power since about a month ago.
I would have survived if I stuck with the original estimate, but the system seemed to be converging on shorter intervals rather than decreased power. Now, at least, I’m scheduled to move into 3x20 SS this week, while my over unders go behind the ten minute mark to 3x12. Seems more effective and productive.
I am curious to hear if I didn’t go down the right path with my decision (but my next AI FTP detection is only a week away anyways - right after my rest week).
I also turned down my FTP for a couple of workouts, but I’m going to see if I crash and burn on my next o/u. I’m a 50 year old man with no racing ambitions, so don’t care if it sets me back a bit - I just find going into the workouts quite stressful!
The funny thing, is it already too high IMO, and it’s predicting (likely won’t get there), to go up another 11w in 25 days again! I keep wondering why the ML has so much faith in me lol.
Anyway, I’m going to let this thing play out and see what happens as I said. My main question is really: has TR pretty much abandoned the training plans in many ways, and truly just going by watts - and the energy systems don’t really matter anymore. Are they more for presentation now, just so people don’t freak out more than they have been? Maybe this is the new frontier Jonathan is alluding to!
Getting the best workouts for each person could very well have nothing to do with staying in certain zones anymore I suppose. Again, I’m cool with that, but would like to hear Jonathan’s view on this.
It does read raw watts and it understands your ability to maintain those watts for specific durations, and it can independently manage that for each zone.
As @Jesse_Vernon1 indicated, the plan framework does decide the zone/profile/duration you’ll do, and TrainerRoad AI runs the simulations to find the best workout within that zone/profile/duration. If it can’t find something appropriate, it will alter the duration, and if excess fatigue is present, it may adjust to Endurance or rest.
But I can’t see a scenario where it would turn a lower zone into an upper zone.
Looking at your workouts since Jan 9, I see 3 Sweet Spot workouts and 3 Threshold Workouts, and only 2 out of the 6 are rated harder than predicted.
Jan 13: 5x10 SS at 266 (You nailed the power and even went +10w on the last interval, and you also rode 20-30w over target on your rest intervals. It was predicted to be Hard, but you rated it Very Hard, but I think I’d expect it to be harder when you’re riding above target for rest intervals and +10w on the last set. Those adjustments are totally fine, but worth noting that the model isn’t prescribing or predicting that.)
Jan 20: 6x8min SS @278-260FTP (You rated this one Hard and it was predicted to be Hard. You rode 20-30w over target during rests again and rode over target on the last set by 8w.)
Jan 27: 5x10min SS @278 (You rated this one Very Hard and it was predicted to be Hard. You rode 20-30w over target during rests again. For comparison, a week earlier you did 3x9 OUs averaging 291-293 and rated that Hard.)
Jan 30: 3x9min SS @281/311 (You rated this one Hard and it was predicted to be Hard. You rode 20-30w over target during rests again and about 10w over target for the Endurance block after the last interval.)
This looks pretty well calibrated to me? 2 out of the 6 workouts were harder than predicted. Assuming you were hitting your targets during rest intervals, they may have felt easier?
This is also a very okay option! You’re welcome to just manually set your FTP. This will make it so you can’t get an FTP Prediction since it would be comparing apples to oranges, but the model behind TrainerRoad AI will still do it’s best to give you the right workouts.
Possibly! We have it internally, but we need to do more testing on it to be sure that not only it works well, but also that we can clearly communicate the implications of that data.
I am still wondering if you couldn’t use the system exactly as it’s used now to prescribe workouts, but keep the AI FTP value internal and just show a 40 minute estimate as FTP to the user while still picking the watt by watt same workouts like now when the AI FTP value is visible to the user.
Maybe that could have led to less confusion? Did you test this internally? But I assume you deliberately decided to show the AI FTP value to the user because of the paradigm shift you pointed out:
Thanks for your post on this forum, it helps to understand the thinking behind all this.
One last question: how do you show predicted difficulty for new users or users with minimal data? Do you blend this with the average predicted difficulty of all users until you have enough RPE data from the individual user? Or don’t you show predicted difficulty for users of which you don’t have enough RPE data.
We probably could, but not sure that would be any more clear for athletes?
It’s kinda wild, because when you look at data (performance data, testimonials, feedback on social channels, company metrics, etc.) this is a SUPER popular release and athletes are loving it. But the forum demonstrates the opposite of that. We’ve never seen such a polarized response like this before from the forum, but the confusion only represents a very, very small portion of athletes.
Our forum users are usually our toughest critics and they are also the ones who tend to more closely represent the power-user phenotype, so we take the feedback seriously and find a ton of value in it.
That said, FTP is a hot topic with lots of opinion surrounding it, and forum users tend to be more opinionated, so I think this just strikes a chord.
We just show the individual’s data. Usually a lack of data results in more noise / less certainty, but it doesn’t take long for athletes to get locked in. I’m always impressed to see it happen even after 1 or 2 workouts. Super cool!
Some are the toughest critics and some are the toughest lovers and some are both at the same time !!
Good to hear and Yes. I made this survey (which you marked as solution ) that showed no matter what the AI FTP value is (in relation to old FTP or 60min or whatever) vast majority of users acknowledged that the prescribed training was about right!
It’s just this edge cases that us power users sometimes have a hard time wrapping our head around like this:
That’s pretty wild for this user to see 352 AI FTP with 20min PB 326. Maybe have a look there.
I like your deep dives into those user profiles to understand more, which often turn out that the user only told half the story or gave some inconsistent RPE and I was just shaking my head and did not envy you and the support team. (But that does not seem to be the case for that user … as he showed PDC and told about very hard RPE ratings)
And looking forward to seeing the new UI that Nate showed that helps us better understand the power of the workouts vs 6 week PB or so.
I appreciate you going through the workouts. Like I said in my original post, I have completed all my workouts and haven’t failed any. I do think there is a difference between being able to complete the workout, opposed to the workout being in the correct zone during the working sets. I’m also totally fine just doing the watts and see where I end up at this point.
Before the AIFTP, my detected FTP always felt bang on, and the SS workouts given to me felt like SS - for me I would rate them at moderate and felt controlled, and the o/u workouts felt like unders on the under portion where I was actually recovering before hitting the overs again. These workouts have not felt like that. I’m really seeing a difference between the RPE on the new system compared to the old for a given zone.
I agree that I could do my recovery intervals a little lower, but that won’t significant affect the efforts in the working sets IMO.
The other consequence is the z2 rides are much more challenging, which I actually like - I always felt they were too easy before.
All this to say, my plan feels like one longer threshold session, and one over / at threshold session. This may actually be a good thing as I haven’t done intervals this hard during base before. I’m def not doing a proper SS and o/u session though - at least not how any coach I’ve had in the past has described them. I’m just coming off a rest week, and my AI FTP is 5w higher now, wish me luck! I may even have to stick to the proper recovery watts between sets!
Edit to add: what is the ML seeing that it thinks I could add another 11w to my AIFTP at the end of this block? I know that won’t happen, but can you reveal in a general way what the logic is to come up with these predicated watts?
@Jonathan thanks for the clarification(s). What I’m not understanding is why then TR has deprecated the old AI FTP detection. Personally I found that it was a good estimation of my own FTP. It was much more convenient than doing the TR Ramp Test or other testing methods.
And biggest promoters! Devotion is a double-edged sword.
Thank you for the analysis of Bones’s January. I’m not going to say “do mine” because I already did it at length here:
So, if you need to see failed workouts to be convinced, you’re welcome to have a look at my calendar. I can follow up on my analysis on Reinstein-2 done in the linked post with an example from 05.02 - my last OU workout done with manual FTP setting - Emerson-2. The OU watts (Emerson-2 335/371) are mostly the same (Reinstein-2 337/373), but the duration of the OU block on Emerson in 10min vs the 12min on Reinstein. So I finished an objectively easier workout, rating it very hard, after an extra 10 days of training, meaning that I was always expected to fail the Reinstein-2 10 days prior. I’ve since looked at HR data, too, (before and after the new AIFTP) which strengthened my conviction that the AIFTP is messing up my zones.
It turns out that I am just too sensitive to intensity. Other people manage to complete such workouts, but that does not mean that they are appropriate. 10 or 12 minutes at threshold is supposed to be achievable, but varying the intensity within the interval does not make it an OU automatically, if there’s no semblance of recovery, yet it doesn’t necessarily make it a failed workout, too. Just renders it inappropriate in my opinion.
All being said, I understand that I’m in the minority, but unlike Bones, I’m not willing to crash and burn with the new AIFTP for the sake of science (I was for one training block, now I’m out of time). Using manual FTP is a solution, but I’d like it to be a short-term one. The minority of users thinking that the new AIFTP is too high and is rendering their workouts too intense is not negligible as per Rizzi’s poll:
I still believe there’s improvement to be made in these cases and I’m willing to bet my house that you, yourself, would’ve told me “your workouts are too intense, maybe you need to reduce your FTP” if I had sent my January 2026 training block to the AACC podcast back in Jan 2025, complaining about failed workouts.
I am very much an edge case as a user anyway. I don’t follow any TR plan precisely, I do a lot of long endurance using “Free Ride”, and also go to long cross country ski sessions. I like to pick suitable TR workouts based on their difficulty rating and my long experience from TR workouts and endurance training in general. For my use case, AI FTP before the update worked amazingly well.
I mostly struggle with the fact that the legacy AI FTP gave very accurate numbers for me, even though I train a lot outdoors and didn’t follow plans properly. It literally gave me exact same FTP value compared to 20min FTP test, just hours before the test.
It would be interesting to find out what is the reason behind my obviously too high new FTP AI estimate.
Hi, first thank you for a great product…I have been using it for many years, but I admit, properly not as strict as some. My challenge is the new AI FTP detection, The old one had me on 415W, and the new took me down to 400W and then gave me sessions that were way to easy. Took a ramp test that put me at 425W and settled at 420W, and it works…though I am now at Sweetspot 8.8 and Threshold 5.5, so might need to go to 425 soon. But the AI keep wanting me to be at 400W…why, this is not a new scale thing, this is just wrong…..It also want me to start my Build Phase with VO2MAX 10.0 (+5.5) - which is impossible for me to do. Is there a trick to tweak it? Kind regards
There have been suspicions that 400W is a maximum value for the current AIFTP. Nobody has reported a higher prediction/detection yet. I guess you should go manual once you’re over 400.
Hey @John_Barclay I would also agree that a 2 watt difference is only minimal, hehe. But it does make sense for your training between AI FTP Detections. You were sick for a good chunk and bounced back the last week and were able to fully complete 3 interval workouts which helped with your FTP.
All that said, the new AI FTP Prediction looks promising! However, I see that on Feb 9th you’re already feeling Sick again. So I’d just be mindful and listen to your body first before chasing watts
I can confirm there is no maximum value set in AI FTP Detection. Predicted FTPs above that are certainly going to be quite rare, but human physiology (not the AI) is to blame
!!
