I can kind of see it more for you if you have never been better - it could see a higher ceiling. I agree 3-5% per month is madness when you’re well past the newbie gains. For me, I’m mostly trying to slow the erosion down!
Anyway, I’m a little disappointed in the TR responses. Jonathan analyzed my workouts and placed the most optimistic lens on me completing the workouts to say, “looks fine to me”. As I stated, just because I was able to complete them, does not mean they were in the intended zones.
I believe folks like us are a huge asset to them for feedback. We are very experienced and have been training for a long time. Like I said in a previous post, nobody wants their FTP (or AIFTP) higher than I want it! For me to step back from that number, it’s not becuase I’m mentally weak or something. Believe me, I know how to suffer and know what my zones should feel like given the effort.
I will continue to give my feedback on the manually adjusted FTP going forward - even if they appear dismissive of my feedback.
I’m not as experienced when it comes to cycling and I start questioning these points about myself. I do think I know my body quite well though. My belief is that they are reluctant to post anything outside of the “most optimistic lens” lest it explodes completely.
Same here, if necessary, but it shouldn’t be. The workout selector has proven to be great with the appropriate context.
AI FTP is a number chosen where you will complete a level 3 TH workout at the expected level of difficulty. I don’t think we’re going to get more than that because that’s the “secret sauce”.
I think that it primarily sees that we’ve been maximizing time in zone and pushing TTE when it thinks we’d be better served with some shorter, more intense intervals.
After completing a grueling “Joe Devel +3” today, for which I strangely received 6.3 / 5.9 for VO2max and 6.9 / 6.3 for Anaerobic (not sure why, most intervals were below target wattage in ERG. I wonder if these on/off workouts expect you to perform “badly” due to the short durations ), the FTP prediction panicked :
I don’t get why they can’t share more. Isn’t their secret sauce the data set they have tied to their workouts to calibrate it? I guess Strava has a similar data set, but any competitor wouldn’t be able to encroach on it without similar data set.
Sometimes, when Reading all this kinda harsh comments I feel pity for the TR staff who worked hard and had success with this great update.
But then, I remember that TR is a business whose leaders took the business decision to try this marketing stunt and use the marketing rally on years end. Nobody forced them. Leaders made their decision. I guess this is a big part of the reason for the negative sentiment. Not that these “little group of very confused power users”
Do the new tools give more insight…don’t the picked workouts vs last 6 weeks look appropriate? Any other factor involved like different reading power meter etc.?
Manual FTP seems the way for you.
If I were you I’d rate those workouts All out if just to get the “why” dialogue so I could say “intensity” and thus force workout selection to have mercy and to force AI FTP towards your manual FTP. I guess if that would be reached you’d have happy times again.
The tools are very good - I really like this added feature- before my latest workout, I did check with the tool and it really didn’t look too bad, but 2 of the 10min efforts really pushed me over the edge - they were at 94% each.
Like I’ve said before, being able to complete the workout is not the issue (though I really quesioned if I was gonna make on this one), it’s more they aren’t in the right zone.
This will put it in better perspective:
My AIFTP right now is: 301w
If I’m honest with myself, how long could I hold that for maxed out: I’d say 10 minutes or so
Now, use that as my capability, 94% of AIFTP for 10m SS is: 281w.
If I could hold that 301 for say 25-30m I’d say that looks about right, but not when I can only hold it for 10m. I am clearly in threshold territory here.
My next o/u workout this week (before I lowered my FTP) was 3x9, with the unders at 286 and overs at 317w. Now we really have a problem, those 286w under ain’t under!
I’m really curious how my lowered FTP goes. Using the tool, the workouts are technically almost as hard, but trade the intensity for longer duration during the working sets. This should work better for me. Building higher TiZ at the correct intensity makes more sense to me - the TR team is welcome to disagree.
I could do this, but I’d like to try this first for a couple of weeks. I suppose I could just go back and rate them all-out and see what happens.
I have been thinking something similar, are the zones correct? My FTP just got a 5% bump from 298w to 313w and the first workout was 3x8 minutes at 99% of that (~310w). I finished it but found it really hard. I am still sticking with the AI FTP as I haven’t failed any of my workouts yet with it but it for sure felt harder than 99%. Im scared for the next threshold workout
In 23 days its predicted to give me another 6% bump to 332! god knows how those workouts are going to be. I think the highest FTP I have seen is 309 and that was 5 year ago.
I think my FTP is more likely around 300 as I recently did a max effort on Serenity climb in Gran Canaria and that was 295w for 33 minutes. I probably wasn’t totally fresh but it was a max effort.
I’m not new to cycling so it’s hard to believe I could get an 11% increase in 2 months.
I’d like to hear the advice from @empiricalcycling here because IIRC he said that if you are scared of threshold workout it’s sign you’re training too hard (maybe I confuse him with another podcast). Whatever my advice would be to not be shy and set a manual FTP instead of sticking to an AI FTP that gives you workouts you’re scared of.
I don’t know where to put this so I’ll just try it here. I’ll probably be flagged for not being “excellent” but at this point I am so frustrated that after being here for 13 years, I think I may need to move on. I started my 2026 training season on Dec 1, 2025. I used plan builder and carefully put together a set of workouts that should get me at least to my 2025 FTP. The results have been dismal, I have had to delete and restart my training plan 4 times because this AI thing just can’t get anything right. The most recent example, I did one moderate SS and two EASY days and the AI said I needed a rest day when I should be doing O/U’s, so now I’ve had to shut down the auto change feature and now all my days coming up are warnings and it’s not showing any workouts for the future, just generic rides. At this point all I can only surmise is that they are getting my max HR wrong. Through testing and racing, I know that my current max HR is ~180. I did some quick calculations and I believe that TR is using the old saw of “max HR is 220-age”. This would place my max HR (according to TR) at 155 which is actually low tempo for me. I don’t know what else to do at this time, TR keeps telling me I’m not being “excellent” but this is a two way road. TR not being responsive to mine and all the other complaints about the new AI system is not being “excellent” either. You can search and see all the other posts I have put in about this problem and the lack of any serious consideration from TR. I am 65, every workout counts, but when TR says my gain after months of training will be -4 watts, what is the point, what am I paying for? The old system worked great and by using it, I maintained my FTP for years when I should have been decreasing simply due to aging out. I can’t have even ONE unproductive workout, much less an entire season at this late stage of my life.
It may help to be explicit about what it didn’t get right?
Can you share your calendars or screenshots of the weeks leading up to the red day after SS and two easy? Was the structure shifted from a typical week?
Hey hpvpaul2, I’m sorry your frustrated and I’d love to help.
I want to clear a few things up first. Fatigue Detection as far as red/yellow days go isn’t AI based, it’s a TSS based formula. You hit a threshold that triggered a yellow day.
You had:
Mon - 103 TSS + Strength Training
Tues - 47 TSS
Wed - 73 TSS + Strength Training
So you say two easy days, but there’s a difference between 47 TSS vs 73 TSS + Strength training
I’m not sure why you would assume that. We are using a machine learning transformer‑based sequence model that captures temporal relationships in activities. It has multiple input heads for various types of input data.
We are not doing max HR = 220-age.
We predict that you’ll have a 2 watt increase, not a -4.
I think you’re in good faith with your complaints. I’ve DMed you my personal cell number. Feel free to text me if you get into another problem and Ill jump on it or answer any questions.
Looking at your career it seems like you’re doing great! You’re doing less volume than before and increasing your fitness!
I have also been wondering this. My FTP decreased drastically with the new detection. I have wondered if the model “wants” that I increase more time in zone and with lower intensity instead.
I am sorry to hear you are having a bad experience. I had a pretty big bump when I went live as well, from 298 to 315 and I was a bit concerned but have been making progress with the adapted plan and seeing how it adjusts as my performance and RPE adjust. My over and under progression has been really positive, at first it calibrated where I was, then it upped the watts in the overs and then went down in watts but extended the TIZ. In this block I am only a handful of watts away from lifetime PBs. It has tought me that I absolutely need to be on the ball (nutrition, sleep, not overdoing endurance) to nail these workouts.
One thing that I found interesting listening to the TR stuff when this all launched was that they have seen through the model that training at higher watts for shorter periods increases adaptation more than lower watts for longer. I am not sure if this is something they have seen across all athletes or not but it is consistent with what humans have learned in the past from AI’s trained either on specific data sets (like the new antibacterial Halicin) or it does its own learning (Chess and America’s Cup) but in each case we have learned something new that wasn’t previously considered.
That said I totally get time is precious especially if you have an event coming up. It is easier to go back to what has worked for you in the past.
), the FTP prediction panicked 
