yeah - that makes no sense for the AI to drop FTP prediction based on that. I do love TR, so I hope they read all this feedback and make some updates. Good idea to bump up w/o’s by 1 watt each time.
4 Workouts ahead to the AIFTP Detection, the prediction went down from 3.4% to 1.9%. But I’m not upset about the numbers. I know it is a simulation based on ML learning and I think, the more accurate data, means since the implementation on AIFTP-Prediction, it will get, the more precise it will be.
Anyway, for me it is the prediction a cool tool to motivate me to be on track. Since the update I was doing all workouts without failing, which was in the past not always the case and therefore it seems to me, that the AIFTP and the adaptions on the plan are good. Despite the number are going up and down, the constant structured training will make me faster, hopefully ;-).
Just a little update. After yesterday’s Endurance workout, the prediction was still at 6% (281 W). Today (no training scheduled), the prediction went down to 4.5% (277 W). I’m currently in the rest week of a Base phase of the 12-week “Mountain Bike - Increase FTP” plan.
Being 1 watt off isn’t going to affect how we classify your workout. What do you mean when you say “but now with AI it’s reducing my results upon completion of the workout?”
Yeah as soon as you can 100% turn off Prediction it getting turned off, I’m just testing it for a few weeks.
Right now I’m doing a short CX endurance plan just to fill in the time until March when my season plan begins. I’m on the four week Build Phase now.
At the beginning of the phase FTP Prediction had me at 242 FTP, I though cool this would be where I left off last season before my break.
Now it looks like Prediction likes it when you rate everything Hard, to bad for me that VO2 and Anaerobic workouts are my weakness (Yet I enjoy doing CX races 
).
From years of training even though I know that FTP Prediction will take a hit I had two workouts that I had to rate Very Hard, one was so borderline that it was the Chad Chart to the rescue.
Now I also know there’s no way that if for the rest of my 11 workouts I where to rate everyone as Hard that I would only get an 1 watt increase.
Well it is what it is, I know as the season progress things start working itself out, I guess I rather have it Out of sight, out of mind.
1 Like
You can already turn off AiFTP Prediction. Do you mean detection? Not sure why you would want to turn off detection.
2 Likes
I turned it off. I wish you could get rid of the widget too but that’‘s ok.
I’ll peek at it when I get my next AIFTP detection then turn it back off.
1 Like
Spot on, OP. I’m 26 days in and have realized that obsessing over the ‘black box’ daily logic is a recipe for frustration. It’s a 28-day forecast, not a real-time merit badge. I’ve seen my own prediction tick down from minor calendar changes, but I’m choosing to look at the big picture instead of the one-off anomalies. If you’re looking for a platform that prioritizes long-term physiological trends over daily ego boosts, this new system is exactly that. Staying the course is the hardest part, but likely the most rewarding.
9 Likes
After my AI detection today I’d say the prediction is exactly the detection (n=1). The two yield same numbers, whether intentionally or coincidentally. So better watch it as you train/adjust the plans.
Ignore the WL levels being equal - they’ve stated in other threads + on the podcasts that the AI evaluates workouts in a much more nuanced way than WLs. My guess is that despite an equal WL, Bounty would provide less stimulus for you than Sonoran based on some combo of interval lengths, intensity, and rest lengths. IMO, you can start ignoring workout levels.
Really appreciate @aydraper’s perspective here, this matches how I’ve been thinking/using TR after the update.
I think a lot of us are applying way too much false precision to FTP. A 1–2% swing in an estimate just isn’t meaningful compared to things like sleep, fueling, or life stress. But if you check the prediction every day, it feels meaningful and can mess with your head. It reminds me of checking your 401k daily… you can end up reacting to noise instead of just consistently contributing and letting compounding do its thing.
Same with trying to “game” the system to maximize the predicted number at the end of a block. If you’re going to optimize anything, optimize consistency. If you feel like you have to select alternates, pick the ones that help you show up, complete the work, stacking solid weeks over time.
For context, my AIFTP is at 294 and it’s estimating 306 soon. Historically I’d be thrilled with even a +4W bump during a base block (I don’t think I’ve ever seen a 12 watt bump in a block). If that estimate drops a bunch between now and then, I honestly don’t care. My power curve is steadily improving and I feel far from burnt out, exactly what I’d hope for in a training plan.
4 Likes
Funny getting the exact same prediction.
It’s funny on the podcast this week they were talking about weight loss and how it’d be great if there was a scale that reported your moving average. They may want to consider something like that for the AI Prediction value. Or just a range that tightens over time.
1 Like
2 Likes
I do not race and work out only 3-4 months in winter. Additionally, I don’t do much sports in the Autumn because of work, and in Summer I do not ride my bike too much, doing lots of hillwalking (at least 3 weeks in July and August) in the Alps. So I expect at least 10 W per month improvement with a reasonably good training plan (and I was able to achieve this using a stock Zwift plan without any AI). I want to get as big an improvement in my FTP as possible during the winter months before I start riding outside again - that’s my reason to use TR.
Not sure if this is the right thread because there are so many similar ones right now, but anybody else keep getting the same workout given over and over? I did Antelope +2 (SweetSpot 5.5) on Tuesday, rated it hard, it then changed my workout today to Antelope (SweetSpot 5.0) which was easier so I rated it Moderate. And then it gives me Antelope again next week. And also thinks there’s a 56% chance of it being hard. Huh? I just did it today and it was moderate. It was actually a little too easy for my liking. Why do I keep getting the same workouts?
1 Like
Ok, now I really don’t understand the AI workout selector. So I do Antelope +2 (5x10 @94%) and rate it Hard. AI gives me Antelope (5x1@88-94 @88-94%) next. Ok it doesn’t want me to have Hard workouts. I finish Antelope and rate it Moderate, definitely a good bit easier than Antelope +2 not surprisingly. Literally wishing it were harder. But now it changes and gives me Antelope -2 for my next workout, even lower level. Huh? Why would you contiunually give me easier and easier workouts as the block progresses? I’m doing the workouts just fine. I’m not rating them Very Hard or All Out. Why does AI keep lowering the difficulty?
Edit: My SweetSpot level is 6.2. AI progression has been SweetSpot 5.5 → SweetSpot 5.0 → SweetSpot 4.4. Not getting any longer. Intervals are getting shorter and easier. What gives?
1 Like
My experience only, YMMV. I’ve noticed sweet spot workouts dialing back a bit for weeks with a bigger threshold progression. How have your other zones been progressing/feeling?
1 Like
I skipped an entire Sweetspot workout this week because I had a cold, and the effect on my AIFTP Prediction was … precisely zero. 
I guess if you’re also doing Threshold at a vaguely sensible level, Sweetspot isn’t really progressing you in that block.
To that end, it’s entirely possible that the Sweetspot prescription from the AI is there to keep enough volume/kJ to support the hard work and no more in order to make sure you’re reasonably fresh.
Or something.
No. I actually only have 1 Threshold workout in this entire block. And it’s a 2.8 even though my Threshold level is currently 3.3. I’m so confused.
Same here with Antelope as well. I asked for a 90min SweetSpot yesterday and got Antelope +2. I rated as hard. I programmed a 60 min SweetSpot workout next Thursday (that got changed to 90min), and it is again Antelope +2 that got selected by AI. So same duration, same workout. The workout library is so huge, I don’t understand why AI gives me the same SS workout twice