Yeah, another thread on this, but I think I’ve got some insights from my engineer’s brain that may help others, and what good is the autism if it doesn’t help in particularly niche and nerdy situations like this?
Caveats:
-I don’t work for TR so I don’t know this is how it works, but I have some I think good guesses albeit dumbed down from reality.
-I don’t have any skin in the game besides wanting the system to train me so I don’t have to think about it (gives me more time to think about how it trains me).
-I’m not an exercise physiologist or coach or trainer or anything besides a scientific-minded layman.
-I actually don’t care about FTP, I only care about good workouts, but I understand why some people care (and in many cases care a bit too much lol)
AI Workouts - Excellent
I’ve seen a lot of comments here and on Reddit that indicate their AI selected workouts have been really good despite the AI FTP changing their supposed FTP. My workouts have been excellent. The intensity always seems to be correct for my hard workouts. This makes sense because the workout AI doesn’t use FTP to select a workout as they’ve straight up said a bunch of times, and I believe them.
Based on interpreting some things said on the podcast and some of the responses to others on this forum by TR staff, I’m pretty sure the workout selector AI looks at something like your 90 day power curve and finds workouts that are above that power curve in two domains - VO2 Max and Threshold. For VO2 it finds workout with a power curve above your 90 day curve particularly in the sub-4 min domain (maybe up to 10 mins), and for Threshold it finds workouts where the power curve is above your 90 day curve in the 4-min to 60-min (maybe fewer, like 40 or 45 mins) region.
From that subset, the workout AI will pick a workout that meets your time demands, and depending on how you rated your previous workout on the 90 day curve (moderate, hard, v hard…) it selects for a workout curve closer to (very hard) or further above (moderate) the 90 day curve.
I think this system works extremely well for me, and seems it works very well for most from the comments. And the methodology makes intuitive sense to me based on progressive overload principles, and backs up why most have been saying their workouts have been good despite wildly varying feelings about their AI FTPs.
AI Predictor and Post-Workout Survey Interaction- Room for Improvement
The predictor seems to be kind of sensitive to the end of workout questionnaire, but I do think it’s ultimately self-correcting. I workout in Zwift and their end of workout rating is 1-10, but they don’t actually correlate very well to TR’s qualitative assessment (Post-Workout Surveys – TrainerRoad).
I find a 7 or 8 RPE typically has at least a rep in reserve, so it is “hard” and not “very hard” per TR, but it is automatically logged as Very Hard. I think it doesn’t matter too much long term, and actually might be better for long-term growth as it’s more measured progress, but something to watch for and could leave some short-term gains on the table.
This also could kind of hurt people in the short-term if they see this and intentionally misrepresent their effort levels to impress the FTP predictor (guarantee some are this insecure). However, I think it will just lead to getting spanked on a workout from the workout selector AI as it will give you a workout with a power curve that’s a big bump above your 90 day curve.
AI FTP - Room for Improvement
Per the podcast and things I’ve seen posted by TR, your AI FTP is based on a Threshold Level 3 workout. I interpret this to mean that a Threshold 3 workout using the AIFTP is guessed by the Workout AI (based on your 90 day curve) to be hard or very hard with a fairly low chance of failure, and based on the workouts in my calendar that seems to be approximately the case.
I think revising your FTP in a manner similar to this actually could work quite well. It makes sense to me that a 60 or 75 min threshold workout should be able to provide some progressive overload, so increasing FTP and reducing the work durations makes sense to me.
However, it is my lay opinion that the threshold level 3 workout structures are actually just too easy (I’m thinking Fang Mountain -2, eg) resulting in an overestimated FTP. I have much more background in running, and the threshold structures executed in running there are more like Threshold 4 to 4.3 (eg Stromlo 4x8mins with 4 min rest) workouts in TR. As is currently, Stromlo sits just below my 90 day curve but after my anticipated AI FTP bump, the AI reckons that Stromlo will be too hard for me. I could be mistaken because cycling and running aren’t 1:1, but I think the Stromlo or San Pedro structures would be a fantastic benchmark/bread-and-butter Threshold workout. I think the workout AI is correct, that at my predicted AI FTP, these workouts will be too hard, but this leads me to believe it’s the AI FTP that is too high.
As currently constructed, I think the threshold development and VO2 development will be fine regardless, because of the previously discussed way the workout AI selects workouts means that you’re bumping up your power curve in that domain with the easier workout structures. However, the consequence of the slight inflation to FTP is that workouts below your power curve, eg the sweet spot and endurance domains could reasonably come in kinda hot, particularly for those pushing lower watts as each “zone” shrinks.
TR touts the tempo and sweet spot zones as a great area for growth, and I won’t disagree, and I understand that physiology happens in a gradient or spectrum and it isn’t light switches for the most part, but I think overestimating FTP for someone working a lot of SS could be maybe the worst scenario. I don’t think the low tempo/endurance transition matters as much, so those operating more polarized it’s less important.
Anyway, I think benching off something like a 4.0 or 4.2 providing a very small power curve bump, instead of basing on a 3.0 may address the sweet spot concern as well as some others people have been having just feeling their FTP is impossibly high. But, I understand you have data and have smarter people than I who this is their field, so I’m not going to push. Like I said up top, I don’t actually care about my FTP, but I do find it interesting.
Closing Remarks
The workout selection has been so, so good, so chapeau for that! I’ll be interested to follow the development of the AI FTP and predictor models. Best of luck and don’t let the haters get you down. I can tell y’all aren’t half-assing this.
Or I could be in newb gains mode as I try real structured training vs Peloton with vibe mountain biking and that is bumping up my relative threshold level
.
