The only question I can’t find an answer to I got a 7.9% increase myself.
I understand the whole it’s a training number but I’ve already been doing intervals 10/20 watts higher than I was last week.
I went ahead and changed all my numbers to match the new number in Garmin and Strava otherwise all my zones would of been +1
One thing I have noticed is my heart rate now tracks equally to each zone so my only guess is my old FTP was underestimated and the new one is closer to what it actually is.
Unless I’m misunderstanding, workouts/intervals have always been pegged at a percentage of FTP. So sweet spot intervals for me will now fall between 222 & 237 watts, where previously it was 237 to 254 watts. Or to put it another way, easier.
This was discussed at length in the beta feedback thread. My experience with it mirrors that of the majority in the thread (which has > 3,000 posts, so I haven’t read them all):
I don’t have any insight into the internal discussions, but TR is aware of this (since there has been plenty of feedback from folks including myself) and they have decided to continue to label the new power setting as FTP.
Should you care?
If you are interested in getting the right workout, then TR AI is a step change for the better. The new system is so much better and so much more flexible than what came before it, I was concerned I had to temporarily give up these features. Coaches your prescribe a workout type and TR AI selects one of the right difficulty. You can switch to workouts and see the effects in the simulation. You could roll your own training plan by simply dropping workout types in the calendar. This e. g. obviates “scheduled TR now” and other feature requests we have had.
Are they perfect? No, nothing ever is. But in my experience and from reading the feedback thread, for the vast majority of beta testers, it is a significant change for the better. I’m close to having the best fitness I have ever had on the bike, despite training less (4 hours of intervals per week and 0–6 hours of endurance work plus two strength workouts per week).
The pill you need to swallow for now is that TR’s FTP differs in definition from everyone else. Hence, comparing my performance to previous years becomes difficult, as is using TR’s “FTP” as a reference point for pacing an effort or pacing yourself in a race.
Here is what I do:
My speculation is that TR will do dynamically generated workouts next with custom “reference power levels” for each of the zones. These will be absolute power numbers, i. e. for VO2max workouts the standard “reference point” is currently 120 % FTP. That will be replaced by a number that corresponds to your ability. The fraction between your reference threshold power and your reference VO2max power will likely still fall in the 115–125 % ballpark, but it will be adjusted.
On the old system the change in FTP would change your PL/AL/WL to reflect a level you can work at for that new FTP. This is still a thing however the ai is looking at what you have done for power and your HR if you have it and choosing the workouts based on what you have done.
For simplicity sake lets look at it from the levels stand point. If your FTP was 225 and it had you at a level 2 for a zone. You would now be working out at say level 4 for a new ftp of 215. This will do a few things. It will have shorter rests for certain types of worksouts, it will have longer intervals for certain types of workouts, and it might push you to a higher end of the zone. Combine that all and it doesn’t get easier just because the theoretical zone is a lower range. It may be harder even if it has you working at a high end of a zone that overlapped with the low end before where you were working but now with less rest and longer intervals.
I’m incredibly grateful for TrainerRoad. I keep getting better and better. Huge thanks to the entire team for the amazing work!
Great post 
Only bit I’d expand on is this point. I actually found that the system can cope with a very wide range of “FTP”s and continue to prescribe workouts that feel right.
The only slight issue I found was that if you put in a lower “FTP” you can hit workout ceilings. i.e. it doesn’t seem to like prescribing workouts above level 8 in sweetspot so you can end up with repeating level 8 workouts.
Yeah, precisely. In my case I was maxed out on Sweet Spot (I hit PL 10.5, e. g. 3 x 30 minutes at 94 % FTP). I reckon that TR simply moved to target from “best estimate of power at MLSS” to “best power setting for our workout catalog at these PLs”.
Congrats on the new AI platform. Does the AI take into account inside vs outside rides when selecting workouts? For example, I know based on some recent 2x20 min workouts I’ve completed recently that I can hold ~15-20 W higher outside steady state then I can inside. I heard in the podcast how you take into account time of day, sure seems like it would be useful if considered trainer vs outside too?
Thanks for all the hard work!
Any advice on how to adjust training plans for scheduled blood donations? I have very selfishly pushed my next donation until after the AI FTP prediction. When I do eventually donate, though, how do I tell the AI that I need decreased intensity for a week or so?
I’m on a 25 week half IM plan and My 1 month predicted FTP increase is 1W? Is the model reducing all my rides to just maintain because of all the run, swim and strength sessions in my calender. I’m looking to improve FTP significantly this season so 1W a month is not worth it
I get it. Absolute power and durations are kept pretty constant no matter which training FTP is set inside TR. I am open to training under a significantly higher “FTP” than what my lactate threshold is. I don’t care that I don’t see a lactate threshold inside TR anymore. I personally (lactate FTP 308 W, TR AI FTP 342 W) just have three major problems that are not addressed by TR staff so far:
Wanted to say great podcast and it was great to have Nate on again, certainly have missed his perspective in the past. I think you guys did a great job going through the thought process behind the change and explaining all the new features and what they do. Excited about the upcoming threshold level to push dynamic endurance outdoors, small change but very cool.
As a beta user, I want to say to everyone worried about the new updates, new platform, etc. give it a shot and try it out before you freak out. There is a lot of change and I think a lot of it is for the better, and also what a lot of people have requested over the years. Longer rides, etc.
I think there is a lot of missing the forest for the trees when very few people have actually messed around with the new platform. Give it a shot for a week or two before making big rationalizations. The vast majority of the beta users really like the new system.
Also if you are having issues. Post your calendar, post your workouts, provide data. There are edge cases with issues, but there are also a lot of users on here who may be able to help you out and quickly and easily solve your problem.
Curious if you tried the new system and workouts for a couple of weeks. If you weren’t in beta, answer is obviously no. But I’ve been in for a while and over/unders feel like over/unders, not VO2 (I’m not breathing like a fish out of water).
I know this idea is hard for analytical types (including me), but I would strongly recommend accepting the AIFTP and trying the system for month or two and seeing how it goes. I (and many others) are finding it to be spot on and delivering very high training quality.
Can’t answer the question on the Garmin issue…but Nate has said they will eventually be giving users estimated 20 min and 60 power numbers. You could use that to manually update Garmin?
I’ve been a Beta user and all the concerns I read, I would say, let the system do its thing, let it again learn you and it will all be filtered so it gives you the correct numbers and training. One important thing here to say that you for 100% fill you query (and hones) as this is part of the system as a whole.
If I just started asthma treatment how can I best update the models the AI has of us? Had childhood asthma, didn’t seem worth it to take meds in adulthood but realized when pushing hard (high breathing rate and heart rate) my breathing was less then it could be medicated (no actual asthma attack) So now that I am medicated I’m guessing I can do more when at high intensities as I can get more oxygen in. What’s the best way to let the model adapt?
Will TR use more data from outside of cycling? For example the general data from a wearable to tell if someone is more stressed from normal day to day tasks and work (I’m sure lots of users use Garmin wearables and similar devices) or other activity types (I like doing long hikes which wears out my legs for cycling)
Thanks @Nate_Pearson for mentioning in the video that wearing a HR strap in a workout helps with training the model for us. I think this is the first time this has been mentioned outside of it helping to have HR data to be able to make better use of other activities that lacked power data. Seems like this should be better surfaced to encourage the use of heart rate data. I do wonder if you keep track of the source of heart rate data as some sources are more accurate then others (I think fit files store this and easy to tell inside the app based on what device is paired) Guessing there is still no beat by beat timing used to calculate more advanced metrics like alpha 1 dfa
@lbosch
I think your concerns are entirely reasonable, but they have been considered by TR’s engineers and have not been an issue for me training with TR AI.
I recommend you try it out, accepting the default behavior. If you are bothered that TR’s FTP no longer corresponds to your FTP, pretend it is called something else. I reckon that the vast majority of people will quickly forget “the old TR” after getting used to it.
In some instances, you have to learn the “new best way”, which may be different, because the platform has new features and capabilities. To give you one specific example, the new platform works best for me when I leave all workouts dynamic for as long as possible. I got some unexpected behavior when I initially wanted to plan my next block and chose e. g. specific, longer workouts rather than changing the duration of the dynamically selected workout.
All of your issues have been raised by beta testers and yours and many more have been discussed prior to even giving beta testers access. In my interaction, TR’s engineers have been great during the beta test, they came across as thoughtful, smart and professional, and they really care about the product.
I don’t think you have to worry about that. In my case, I had to do a threshold workout yesterday that went up to 102 % of “FTP”, which meant that I was closer to 104 % of my “true FTP”, i. e. within threshold range in the traditional sense. The same goes for sweet spot and VO2max workouts.
I have not encountered a single instance in my training where the shift in FTP led to a change in the nature of the workout.
In the cases I have seen, TR AI’s “new FTP” upped the FTP to get out of a PL range where there were few workouts for the athlete. For instance, I maxed out the Sweet Spot PL and now I have many more options thanks to my “new FTP”.
This is an issue that remains. I’m on Wahoo’s platform and I don’t think TR’s FTP syncs to Wahoo’s cloud. I have to set it manually. I speculated above that this will be resolved once TR introduces dynamically generated workouts.
Overall, calling it FTP is suboptimal, but I can see pros and cons both ways (picking another name vs. keeping the name, but changing the definition).
How did you rate the RPE survey of that ride? Did you put in “All out”? I can only imagine it has to do with the RPE ratings on your recent hard workouts. If you rate these on the easier side that could definitely have an impact on your given AIFTP. Maybe you can go back and change these and see if something changes. And if you are in doubt, always choose the higher RPE rating in the survey.
Thank you and Great podcast. Question related to the little snippet where you mentioned the large library of workouts. I started a plan under the previous iteration of the models. As I look through the workouts I see the same patterns week after week not in just type of workout but identical workouts. Will the new group of ML models choose more diverse workouts if I build a new plan? Or can I count on the ML simulation to add diversity as it re works the existing plan?
Thanks in advance for your help!
Jason
Except that you missed the part where I wrote that I tried the beta for several months now and it has worked very, very well. I called it a step change in the positive direction. It is eery how accurate the predicted difficulty of workouts has been for me. TR AI was more aggressive giving me seemingly very hard workouts. It also gave me partial credit for workouts I did not complete as prescribed or where I overdelivered.
My point was to assuage concerns that TR has missed anything that glaring or obvious. Even in cases where TR’s dev team made choices different from my suggestions, they seriously weighed our input.
How do you come to the conclusion it doesn’t work? Have you tried it? I have, and it is much better than Adaptive Training and my impression is that this sentiment is shared amongst most beta testers. (The feedback thread has >3.7k posts, so I haven’t read every single one. But I have been using it for longer when there were fewer beta testers.)
Feel free to ask specific questions.
Seems like you may want to read the first post here:
Goes into details about the new ftp