Am I ever off the hook, though? 
The big, often missed point in many of these posts, is that “detected FTP” is also OFF and giving bad, nonsensical results (at least for riders doing all rides including structured intervals outside). “Predicted FTP” problems are the most obvious (so many reporting the same problems), but IMHO the problems with “detected FTP” (eg, dropping FTP after 4/8/12 weeks of consistent execution & performance of structured interval sessions, when you know fitness has improved significantly). This has me on the verge of dropping TR after this new AI release with big (and unaddressed by TR on the podcast) problems, likely will move to FasCat due to loss of trust in the platform
100% agree, I’m in the same position as you. 
So this is the “endurance ride” bug I’m seeing.
I do a hard ride on Saturday and Sunday is a “recommended rest day” but presents the option of an endurance ride. In this case it’s showing a 30 minute endurance ride option.
But when I select the ride it changes it to a longer endurance ride; in this case from 30 to 45 minutes. This one wasn’t as bad as the last one that changed from 30 minutes to 90 minutes.
This makes no sense. I understand rest day is recommended but if you give an option of a recovery ride you wouldn’t expect the system to make it LONGER when you actually choose to do it.
Plus it makes planning may days difficult.
1st pic is before I did anything.
2nd pic is the detail
3rd Pic is after I selected the ride over rest.
Do everything you can to be consistent this month and see what happens. Also, check this out if you haven’t already, has some very clear steers.
It’s been working pretty well for me, although had a brief drop this weekend after my first outdoor ride, but it seems to have corrected itself after another outdoor ride today. Follow the rest recommendations and don’t overdo it, and if you’re rating in line with what the AI thinks you should be rating, you’ll hit the prediction.
It’s hard to keep the discipline, because I really want to push KOMs or just go hard on an easy ride, but I’m trusting the process 
This definitely seems like a bug of sorts. I’ve just created a report and I’ll follow up when I know more.
For the time being, I’d recommend avoiding riding on red days. If you’re ever looking for 30 minute recovery workouts, you can go to the “Browse Workouts” tab and filter for Active Recovery workouts that are shorter than 45 minutes long.
We’ll have recovery workouts plugged into the AI Workout feature soon, too.
Sorry for the trouble!
The thing I disliked the most is that FTP basically changes based on the responses to the post-workout report. I think it’s wrong because the response is very subjective, while FTP is an objective measurement. I think next year I’ll use the ramp test, maybe once every two months.
Even the suggested workouts for THR often left me perplexed. I did 90-minute workouts because I only had two days off, and often the last 20 minutes were easy in Z2 until March, then suddenly become very hard (probably my ftp was wrong?) I could and would have trained more, like I did two years ago.
In April, I found myself well-trained in VO2 max but behind in THR and Sweetspot.
I love TR but i’ve to fix this two item:
FTP with ramp test
Thr and Sweetspot Workout longer since december.
Just to clarify, are you saying that you do want to use ramp tests to set your FTP?
Also, are you looking for longer workouts or just more sustained intervals in those sweet spot and threshold workouts?
Now it’s too late. I will work on it in October.
I have very mixed feelings about the new TR AI but at this exact moment I’m frustrated, like the OP. Since it’s implementation it’s decremented me 25 watts, nearly 10%! Admittedly work demands have resulted in less consistency than usual, but it hasn’t been that bad.
I feel TR AI over emphasizes the importance of VO2 workouts and does not appropriately account for threshold work. It also has no idea how to handle my one tempo run per week.
It also keeps giving me stupid easy endurance rides, whereas I feel that riding in high Z2 builds sustainable, long-term endurance. Admittedly this does need to be balanced with overall fatigue, but I can do that using Workout Alternates.
Basically I’m beginning to feel that TR tries to focus on ‘fragile’ FTP gains more by ‘pulling from the top’ (eg VO2) instead of long-term gains by ‘pushing up’ from the bottom (eg endurance, tempo, sweet spot, and threshold). I’ve thought this for a while but it seems like it is becoming more obvious with the release of TR AI.
I will continue to trust it for another block. I have edited my plan to switch from the balanced approach to more aggressively build threshold and reduce the emphasis on VO2. To be honest, though, this backward progress has me looking at JOIN again.
What is different with Join’s approach?
I’d stop trusting it. Your training time and effort is too important to waste on inappropriate efforts.
it’s really frustrating. Worse that after several podcasts I’ve not heard any specifics on resolution.
I share the frustration and sorry TR but I cancelled my subscription. On April 20th I renewed my subscription and after one more year of just running I was glad that my predicted ftp wasn’t so low. From 380w at my peak the predicted TR AI ftp had decreased to 317w. Not bad I thought after one year out of the bike. But the best was that TR AI was predicting that I was coming back to 380w in just one month 
. Finally after the first month it predicted 375w and I tried to train trusting this ftp. Give it a try and trust the AI I was advised. Obviously my ego wanted to believe but I see it clear now that it was totally wrong and it only lead me to overreaching and extreme fatigue.
Now I trust much more Coros Evolab (free) 335w prediction but it looks like TR AI doesn’t like much setting manually your ftp. So I am wondering how is going to be TR useful for me in the 10 months before my current subscription ends…and I am afraid that the “money back guaranteed” doesn’t apply to my subscription.
Sorry again TR, your customer support people are great but your publicity is really deceiving
I don’t think it discerns whether we’ve slacked off or not. The outdoor portion is completely broken. I’ve hit my last 3 weeks religiously and nailed all the workouts and rated rpe appropriately. I’ve slid from predicted FTP for 321 to 313. I know @eddie says they are “close” to a fix. I have no doubt they are trying but as future workouts are being degraded due to a bug, it’s beyond frustrating and hard to accept that the software is actually elevating performance.
I wonder if they can’t really parse outdoor workouts in enough detail to make sense of it. As an example, if all they can look at are duration, NP, IF and a power curve for the ride, that might not be enough info compared to an indoor ride where the software can look at every interval and whether you hit the target or not.
Maybe. But if that’s the case, it really puts a dent into the marketing of the product as something that can be taken outdoors.
Agree. It would be a huge problem. If I do an outside solo ride (ie not following a specific TR outside workout) but say I spend a lot of time in threshold I assume TR can sniff that out and hopefully adapt subsequent workouts accordingly. Like maybe adapt a scheduled threshold workout 2 days later to something else.