This might be of interest:
https://physoc.onlinelibrary.wiley.com/doi/epdf/10.1113/JP273480
I experienced similar to you where for a few years I got good gains by just doing SST. That eventually got me to a plateau and needed a different training stimulus to break through it.
This pretty much describes my recent experience too. Hitting hard workouts, doing them well, and seeing little improvement (most recent assessment was to keep ftp the same despite being able to complete level 10.3 sweetspot workouts relatively easily!). I have been doing good threshold and vo2 work as well, so it’s not like it’s only sweet spot.
Thanks for sharing. I’m glad I’m not crazy. I’m pretty sure they changed something with AI FTP detection and it doesn’t make sense.
I don’t feel stuck. I feel much stronger. My feelings don’t correlate with AI FTP detection, so my hypothesis is that they changed it.
100% agree and I’ve noticed it since at least the beginning of last year. IMHO, in an effort to NOT burn people out TR plans and AIFTP have become extremely conservative. This effort is alienating and frustrating those that are pushing hard to try to increase their FTP. Unless you’re newer to structured training, the plans have become irrelevant and frustrating. It appears no matter what volume, PL or structure you do, TR is reluctant to give much of any AIFTP increase. Is AIFTP broken or do TR plans and structure not work? I don’t count the cursory 1-3 watt increases either.
AIFTP was once one of the few features of TR that I was a big proponent of has now become annoying and demotivating. Why aren’t TR plans improving AIFTP even when what is recommended is exceeded?
We haven’t radically changed anything about AI FTP Detection, just for the sake of clarity in this thread.
Does your current FTP feel accurate? I think it’s also important to consider whether or not your detected FTP feels correct now in addition to how it compares to your historical FTP values.
In your upcoming Build Phase, you could try out this workout (Marion), which we recommend to athletes who are questioning if their FTP is set correctly.
If that workout feels about right at your current FTP, then there’s no need to change it. If it feels too easy, then you could bump up your FTP by a few percent. With either result, you’ll also be providing AT/AI FTP Detection with more data it can use for your future detections, which will help get you further dialed into the right workouts.
Anyone who doesn’t think AI FTP accurately reflects their FTP needs only do a 2 × 20 at their AI FTP number to confirm or deny it. RPE of about 8. Maybe 9 for the last couple of minutes of interval two.
You’ll learn something, and get a good workout.
Try a ramp test.
If you doubt that ai ftp is accurate go do one of these tests.
If you have done them a couple times you can probably dial in ftp via RPE in about 10m or so.
Many were saying things like the ramp test or AIFTP over estimated their FTP. Now, at least for me, while it has come down a bit, it seems more in line with what i am probably currently capable of. Just my opinion but whatever changes were made seem to have it dialed in more.
Is AIFTP not constantly using machine learning and adjusting, particularly over time as more and more user’s train with their AIFTP number? Or is it just a static calculation and preforms exactly the same calculations today as when it was released years ago?
I’m partial to the Alpe Du Zwift full send myself. 
All joking aside, I think the combination of AI FTP and Progression Levels really muddies the waters here. I used to not put a whole lot of mental energy into my FTP because it was really just a number that was used to establish my power zones for workouts. I’ve been training with a power meter since 2018 and I don’t think I’ve once ever thought to myself, I’m just going to try and sit at FTP for the entirety of this race.
With that being said, now that we have progression levels, it gets confusing. Like if you’re FTP is 300, but your progression levels around in the 2-5 range the workouts are significantly easier than if you have the same 300 FTP but progression levels in the 6-9 range.
In that context the combination of FTP and PL is really just the TR secret sauce to pick workouts that should be the correct level of challenging. So if they feel too easy, mark that on the surveys, if they feel too hard mark that on the surveys and then TR seems to do a good job adjusting future workouts to the correct level going forwards. I wouldn’t worry too much about what your actual FTP is, but if you want to know, you really gotta get out there and at the very least do a 20 minute test.
You’ve missed the point. The discussion hear is not whether AIFTP is one’s “True FTP” (by whatever your definition is). The advise of trying another method of finding your “True FTP” is just another number to train by, which defeats the purpose of using AIFTP and TR plans. I don’t care per se what my “True FTP” is. I just want a consistent method that reflects my ability at this time that I can compare month over month and year over year. Just like my power meter, I don’t care if its accurate, as long as its consistent.
I need a consistent reliable number to set my training zones, not a pacing strategy for a TT.
You are saying up thread that ai ftp is not seeming to reflect where you think you are accurately.
Doing another form of test that is straightforward to execute and provides a reliable marker of fitness that you can track seems like a solid solution.
Sure it would be ideal if ai ftp was working how you think it should be but seeing as it is not doing another form of testing might be necessary.
I’ve been there and done that, but I thought we had moved on from static methods. I did TR ramp tests for about 4-5 years until AIFTP came along. Then I did several ramp tests along with AIFTP and they were within a couple watts. Great, I never need to do a FTP test again.
So now I’m being told I need to reintroduce a new method. How many times? Is that my FTP method from now on? When do I begin to trust this new method, certainly my first go will not be reflective of my ability. Am I back to doing it every Tuesday after my rest week so I’m rested/tapored? Awesome, I haven’t done that in years.
Up to you.
If you want to use the tr ai method then go that route, I was just providing an alternative if you don’t fully trust it.
I can describe how I and many of my athletes track fitness improvements but it’s nothing special.
What else are we to suggest when athletes are (tacitly or otherwise) saying that the number AI FTP Detection is serving up to them is “wrong” - or at least they’re not confident it’s “right”?
If they’re not going to test (or verify, if people prefer that term) empirically then what’s left? Find another algorithm and feed their ride data into it and see what comes out? Not sure that moves us forward.
If we’re really at the point where a 2 × 20 at an appropriate RPE isn’t a good idea to verify a number with “FTP” in its name … I may just have to give up. 
Now this is a good question. My follow-up would be “… and if it is dynamic, what sort of changes would we expect at population-level and at athlete-level and how would they interact?”
Yes, this is a really good question! If it is AI, then surely it is constantly changing, thus, to say we haven’t radically altered anything about AIFTP wouldn’t be relevant as, surely, the system alters itself?
Oh
And there goes every excuse about being too tired/not rested for an FTP test then 