Same feeling for me. After the new AI came out, I got a +8.9% increase of my FTP (270W - almost the highest FTP I ever had). After this raise I also started to fail some workouts, or rate them ‘all out’ or very difficult, which almost never happened to me before. I found this demotivating. I generally like training / suffering but that was too much. After 28 days, I got a -5.8% decrease to 256W (after a week break from cycling - but with 6 days Xcountry skiing instead). Seems to feel more comfortable. But the AI predicts that I will be back to 270W in 28 days! For me this variability is a sign that the AI has an issue : I agree my fitness can have some ups and downs, but not so much in a few weeks.
And I disagree with the arguments expressed above saying that it’s due to sleep, other trainings, other reasons… This started exactly with the new AI release, and in 4 years of training with TR I have never seen FTP Ups and Downs like this. For example from October 2024 to June 2025 I was constantly between 241W and 246W…
Other sign that something is not working: I did yesterday a sort of ramp test with my doctor to check my heart as I’m 60 years old and I have some tachycardy (which, BTW also restarted after the FTP increase) : the output in the test was almost the same as 6 months ago. My ‘old AI FTP at the time was around 245W.
Almost the same or the same? Was it a sub maximal test? I had to do an echo cardiogram and I didn’t break a sweat to hit the HR they wanted, I’m not sure that I’d see any difference in final speed if my FTP went up 10%.
It was an ‘all out’ effort, like a standard ramp test but with an echo cardiogram done while I was pedalling. I reached 5W more than last time, so ‘nearly’ the same.
Same Here! I know Nate doesn’t like FTP and would like to come up with a new metric but let’s step back. FTP is defined per Coggan and Hunter as the steady power you can sustain for 1 hour. I was always suspect of the AI generated number especially for indoor sessions (I am a firm believer I can generate more sustained power outside than on a trainer). As you experienced with the new AI training I got an automatic increase in FTP when it was released with much bigger projected gained that I realized in successfully completing workouts plans in the past. I was puzzled to say the least since I was already suspect of the FTP number it was giving me. Now my project numbers are going down. If I am honest with myself the numbers are still somewhat higher than what I believe I could hold for 1 hour. In the end I am going to say despite Nate’s dislike for FTP the models and ML they are using now are adjusting based on real work what my real FTP is. I wish I didn’t have to go through the optimistic projection that TR is giving me because it is a mental downer when I fail the WO. In the end I think there is a lot of learning that is still taking place in the model.
This is not true. There is no “absolute” definition and it was definitely not “the steady power you can sustain for 1 hour”. I also wish TR had not called it “AI FTP”, but that’s just because it becomes yet another of many “definitions” of FTP.
From Training and Racing with a Power Meter written by Coggan Hunter and McGregor in the glossary FTP is defined as:
The highest power that a rider can maintain in a quasisteady state without fatiguing for approximately one hour. When power exceeds the FTP, fatigue will occur much sooner, whereas power just below FTP can be maintained considerably longer.
Read what you wrote and what they literally said in that book that they authored. They are not the same, and Coggan would tell you that, as he has done many times in interviews. Ok, I’m out. The internet is full of people pointing this out, so I have no interest in watching it be argued for the 1,000,000th time.
This is from TrainerRoad’s onboarding when I signed back up a few nights ago. Is there any wonder why so many are confused when you have people saying that TR’s FTP isn’t FTP, yet TR themselves use the hour power themselves to define FTP. The whole thing is a bit of a confusing mess.
Agreed … but also it’s so incredibly easy to rationalise. Where people run into problems is when they insist that two different metrics with FTP in the name must have exactly the same definition and value.
If people just let go of that and accept that the two metrics are closely aligned but with differences, it’s easy.
.