Typically hard/very hard. If take pauses it feels very different from an All Out effort (short VO2Max without pause would get me there). The system sees the pauses now anyway, and downrates those workouts significantly.
But despite that everything else seems silly optimistic.
Is there a bug that down rated workouts don’t correctly feed into FTP predictions?
I haven’t heard of this bug - and I do agree that the AI sees the pauses - but I did hear the guidance a few times from TR that if you can’t complete the workout basically exactly as prescribed than it should be All Out. It would be interesting to see how your FTP prediction would respond if you had rated it this way.
Friday confirmed to get the get the bump to 303. After failing a couple 3.9 threshold workouts following the bump to 300, I was dialed back to 2.8 the next week and it was smooth sailing from there. Adequately hard workouts but nothing unreasonable.
Was talking to a fellow TR user buddy who isn’t on the forum this morning and we both agree we’re really enjoying the base phase so far. We’re holding .7-.74 IF for longer than ever before. Excited to see how the build phase goes.
Adding another data point here:
TR Detected prior to AI: 304
AI FTP (Jan 10th): 314 | Prediction: 322
AI FTP (Feb 7th): 311 | Prediction: 332
AI FTP (Mar 7th): 321 | Prediction 335
The drop in February was mainly due to irregular training (life happened) and failed SSp Workout. Last period I tweaked the schedule going to balanced from aggressive approach, and down to 5 from 6 trainings a week. So I guess the learning is: aggressive never really worked for me and more rest helps 
Started with an AI FTP prediction of 268W, after four weeks it bumped me to 276W. Compliance maybe 80 percent because yeah… life, a couple rides got swapped or shortened but the main interval days were done. Funny part is the prediction was pretty close to reality, I did a hard 20 minute effort outside and landed around the same range, so for once the robot was not talking nonsense.
Prediction: +3.1%
Actual: + 2.1%
Dialled in 
Threshold & O/U were not quite right for the first cycle or start of this block (too hard). The prediction settled on 2.1% after the first half and at the same time the system then started giving spot on Threshold workouts 
Next cycle: +1.8% (5w)
It’d fallen to 258w after marking a few over/under workouts as very hard but my first group ride on the power meter equipped saw it rise to 259w. That’s not really representative of general group rides as it was the Cambridgeshire Classic Sportive.
Feb 6th: FTP=274, Prediction= 289 (for March 6)
March 6th: FTP=273 (-1w), Prediction= 289 (for April 3)
From Feb to March I “lost” 1 watt. Compliance on hard workouts was there, and some later Vo2 Max I rated as Moderate. I had a couple more yellow and red days than expected (maybe 2 yellow and 1 red). Not sure, but I think doing more TSS than expected might have actually meant a downgrade. Otherwise it’s a head scratcher, because I felt good.
I hit the detect button yesterday at the end of my rest week. Prior to hitting it, my AI FTP detection for 28 days out was showing 237. I hit the button and it gave me a current FTP prediction of 239. The AI FTP detection then updated to 242 in 28 days.
I don’t know, but it doesn’t really inspire confidence that the AI has everything dialed when it’s predicting one number a month away, then it gives me a number that’s higher right now as a detection rather than a prediction, and then changes its prediction based on my FTP being updated.
I don’t necessarily think its wrong. These numbers all seem plausible or believable at the moment and I’m not going to get hung up on a handful of watts here and there. It just makes me skeptical of the prediction numbers a whole lot more. Its still in beta so here’s hoping TR get it dialed in a little better so it’s more confidence-inspiring rather than head scratching.
March update:
AI FTP Mar 8th = 236W
Current AI prediction (Mar 9th) = 237W
Current Plan = Masters 40k TT, June-ish peak, Overall Balanced
Good compliance this month and I’ve removed the extra riding at the end of my main TR workouts (mostly). Really the key though has been managing recovery, sleep, nutrition to nail the hard stuff.
I’m still very happy with the workout selections and they’ve all been appropriate difficulty - I really surprised myself by completing a WU that I thought was impossible but TR AI begged to differ 
Not too fussed about the single watt or two gains as I’m starting to feel a lot more comfortable at this level - I looked at my upcoming threshold workout (Madison) and thought “yeah totally doable”
Going to be a rough couple of months on the compliance front so I’m looking forward to seeing how AI FTP copes with that.
Another 4 weeks complete:
Starting Detected FTP: 226
Starting Predicted FTP: 236 (+4.4%)
Final Predicted FTP: 231 (+2.2%)
I had an 8 day ski vacation planned right in the 3rd week. It was a 6 hour time zone difference and I was a bit jet lagged for my fourth week of this cycle. So achieving less than the prediction was totally expected.
On detection day I did a long outdoor ride, flat roads and tried to keep RPE steady. It was my first non-TR ride of the year. It was also a good way to estimate my FTP myself. I rode for 2:13 at an average power of 183W, NP 201W and IF 0.87 (using the 231W FTP).
I would say that a 231 W FTP feels about right or slightly low. I’ll roll with the AI FTP because it’s served me well so far.
How high are your Athlete Levels? If your ALs are high and your workout duration are short (less than 1:30), then this behavior makes sense because the AI would have fewer workouts of the right power and duration combinations to use to serve you the right workouts while your set FTP is lower. Then when your set FTP is higher your ALs are lower and it can pick from a much higher number of workouts with smaller differences between them.
So now it can find the Goldilocks workout for you each day rather than have to pick something too fatiguing or not stimulating enough. That’s my understanding.
That could be it I suppose. I don’t know what TR has available for workouts at the various ALs and if I’m limited with what is available (seems like TR should build some workouts to fill gaps if needed), but I’m in the 3s and 4s of the ALs after the FTP update. But it would have lowered me to those levels after the FTP update (you can see these are based on the new FTP).
I didn’t note what workouts it had scheduled for me before the FTP update, but I do recall it suggesting a longer workout rather than 1 hour (maybe for my Wednesday workout this week), and I think it mentioned not having a workout at the 1 hour duration for me or something like that. That’s no longer the case and I have VO2 and SS workouts in level 5s for the next 3 weeks.
I don’t have a screenshot of what my ALs were before the FTP update, but as a point of comparison, my last two workouts before last week’s rest week were “English” at VO2 Max level 6.1 and “Tallac” at Sweet Spot level 5.1 … both posted below.
At any rate, it seems like you may be correct if it didn’t have a good workout (or workouts) to prescribe for me. Maybe I should have accepted the longer duration workout (or workouts) before detecting the new FTP and see if the prediction would have gone up based on that. I don’t recall at this point if it automatically put the longer workout in my calendar or if I had to accept it first.
It depends on zone. An hour is plenty of time for VO2max work because you only need 20 minutes or less time in zone.
But sweetspot quickly runs out of ability to increase time in zone with only an hour session, especially with a warm up and cool down included. Tallac is 3x15 min at 88-94%. There are harder 1 hour SST workouts but they don’t actually get much harder (despite ALs going higher). So there wouldn’t be enough headroom for progressive overload the next 4 weeks unless you had an FTP bump.
Similar issues happen at lower zones of the workout isn’t long enough.
Basically the choices become increase your FTP or increase the length of your training sessions or replace sweetspot training with threshold training (which causes higher fatigue).
Adding my metrics after nearly two months of TR AI:
Aug ‘25: AI Detection: 309 (general recent high)
Jan 1st: Manual FTP: 290 (after 3 months off)
AI FTP
Jan 21st: Detection: 313 + Prediction: 352
Feb 18th: Detection 334 + Prediction 351
Mar 18th: Currently 349 (with only 3 difficult workouts to go, so prob achievable)
Summary:
Maybe something for @Nate_Pearson and the rest of the TR team to consider … Should the AI be able to see possible limitations in a training plan and suggest options for addressing it?
“Hey! We’re seeing that your current training plan may be limiting your gains. We recommend you [Check Volume] to perhaps increase volume for Sweet Spot workouts, [Adapt Training Plan] to change Sweet Spot workouts for Threshold workouts, or [Detect New FTP] because it’s been over 28 days since you checked your FTP and if we lower your Athlete Levels based on a higher FTP that would allow us to select more productive Sweet Spot workouts for you without increasing volume.”
Nice one Eddie and welcome to the forums!
Another 4 weeks complete:
Starting Detected FTP: 293
Starting Predicted FTP: 305
Final Predicted FTP: 298 (1.7%) on 06-Mar, had hernia surgery on 03-Mar, so a couple of days of no riding missed two hard workouts.
Had one Threshold Over/Under that was a fail. The overs were 322 which my diesel engine was not a fan of. I did a number of zwift races instead of Threshold work since I had the surgery coming up and wanted to have some fun, prioritized sweet spot as in my head it feels more durable. If I had moved threshold to my sweet spot on Thursdays, I probably would have been pretty close to the initial number. April 3rd is predicted to be 287 that will be after 3 weeks off the bike and modeling a week of 30 and 45 minute rides and a week building to 90 minutes all easy endurance.
I have a hand surgery on Thursday and will be in a cast for 7 to 10 days after that which will be the rate limiter to riding I think. The hernia surgery probably could have waited but this way it’s one recovery period and I don’t have to worry about another surgery. The hand is take two on a mountain bike crash in September, broke two metacarpals and one decided not to heal.
A low volume rolling road race (I think) training plan started Jan 2nd at 219. Probably pretty low from a fall of run training and almost no time on the bike. This was reflected by a quick change to 238 on the 12th. I promptly got sick and went on vacation washing out the first block, finishing at 239 on Feb 9th.
This last block predicted 261W I think. I finished the month at 260W. Fueled the workouts, didn’t miss workouts and everything felt hard but doable.
New prediction is 281. Curious where it ends up, but tonight’s VO2 workout fit the perfectly hard description.
Real newbie gains, building on a history of running a lot but not really having structure until last fall. The algorithm does seem to expect 100% compliance.
5 Jan Detection: 267w
17 Jan Detection (beta access): 260w
First prediction: 280w
14 Feb Detection: 271w
Made some adjustments to my plan, A race is now an XCO event in mid-April
Prediction: 280w (again)
Detection this Thursday - currently predicting 273w
I competed in 3 XCO races between Feb 8 and Mar 1 so I had to move my workouts around to accommodate for travel, recovery, etc, I’m sure this played a role in my actual FTP not reaching the initial prediction. I feel like my progress has stalled a bit the past 3 weeks but also feels like I’m carrying a lot of fatigue so this week being easy + leaving town for 4 days with the family should help me reset and hit the ground running with specialty phase next week.
Not really concerned with my FTP currently, the XC courses I’m racing between now and mid-May require a lot of anaerobic punch so I’m hoping the specialty phase will sharpen things a bit on that end.
