Just a quick note on that is that when TrainerRoad released progression levels they specifically said that the each level was not scaled the same.
I.e. an “average” to rider doesn’t have all 5s or all 4s or anything like that - so having a higher number for vo2max doesn’t “necessarily” mean you are relatively strong in that area.
I get tested by my coach with her equipment using, I think, some sort of a modified(?) ramp test that is easier than the TR ramp test. Earlier this year, the coach’s first test resulted in an FTP that was roughly 20w higher than TR AI. With tests every few months, both my coach FTP and TR AI FTP both have been rising but the differential between coach FTP and TR AI FTP has been getting smaller and smaller. The differential was only 6w a few weeks ago. My back of the napkin theory is that as I am topping out in my ftp, the delta between my coach’s lab tested ftp and TR AI ftp is getting smaller. Either that or the AI training is getting better.
I’ve run it alongside Xert, which does a rolling/daily ftp adjustment and in my opinion was much more accurate to real world performance but always within 5-10 watts of the TrainerRoad estimates. I’m sure they are analyzing the data in a very similar way but TrainerRoad was ALWAYS higher by 5-10 watts.
Fixed that for you 
I’m still not convinced
Did the Wahoo 4dp a week before I started my training on TR, got a FTP of 274w (End September)
Did the ramp test a week later for TR and got 263w.
Yesterday my AI FTP went up to 274w, so effectively where I was a week prior to that ramp test.
Based on our analysis of thousands of athletes, accounting for their contribution of anaerobic energy to the production of 20 minute power, the average percentage of FTP to 20 minute power is 94% on Xert. Any system that uses a 95% rule or is trained on FTP that is based on this rule, will in general be 1% higher than Xert’s number. This is not a comment on which number is better only a statement of what the difference may be due to.
Note that Xert does not employ a 94% rule in the determination of FTP but assesses the contribution of anaerobic resources to the production of 20 minute power (high intensity energy). This means that your percentage could be higher or lower than 94%. This would then result in a greater difference between Xert’s TP and other systems’ FTP.
I think it’s good for indoor training. I ramp tested twice after AI FTP and it matched.
I can’t say that I believe in outdoor training after this season, so AIFTP probably isn’t a problem for outdoor so much as outdoor training is flawed itself.
Given FTP is really now just a training number I don’t think accuracy matters nearly as much as it did five to ten years ago.
If your entire self-worth is not tied to your FTP, are you even serious about training, bro?
Ah, the internet - where satire goes to die. 
Its the most reliable FTP estimate imo. Garmin, xert, strava, icu, join etc I would never trust.
I have a pretty good feel for my current FTP and TR has never really surprised me.
Always appears spot on when combined with AT, PL and training plans.
It’s good. Between AI FTP and the Progression Levels, the prescribed workouts are on point. Certainly much better than the Ramp Test ever was for me.
I feel pretty good about it. Similar numbers to a 20 minute sweat test without all the aggro of that.
I’m feeling pretty good about it because it went up today 
Good point. My AIFTP seems to be broken, stuck at the same level all year. Clearly rubbish.
I think its on par with my old 20mins tests but I haven’t done one (a 20mins test) for a while. Certainly for me its more accurate than the ramp test. It generally feels right, maybe occasionally a little too high on long term threshold and above but a little too low on shorter VO2max stuff, so I’ve accepted AI FTP D as a training metric.
There isn’t a fix relationship between 20 min power and 30 second power. You’d hope the AI model is also trying to estimate those values and not just doing a standard multiplier of the AI FTP.
I’m not an FTP expert by any means. So here is my anecdotal analysis.
-Trainerroad ramp test: Workouts afterward always felt about right. But the day always seemed wasted - doing a ramp test and being too tired to add more time to the trainer.
-Traditional 20-minute and 8-minute tests - never did well with these pacing, especially the 20-minute - ESPECIALLY when first getting into indoor training. How would I know what pace to set? Never accurate. Even when I learned better pacing, the subsequent workouts that followed seemed too easy.
-Zwift Ramp test. Real ego boost. Felt good for a week then manually would drop my FTP.
-AI detection - never had an issue. Sometimes it seems low sometimes or a bit high other times, but workouts are always appropriate. This feature has been the most consistent measure of FTP for me if the measure is to have appropriate workouts for the next 4-6 weeks.
Keep in mind that that FTP is an operationally defined variable that is simply a benchmark used to develop training plans and measure progress. If your TR workouts are consistently at the proper level of difficulty and you are making progress, your FTP is set properly.
FWIW, my AI TR FTP, intervals.icu, garmin, and on the road ramp, 5, 8, and 20 minute tests are all consistently within a reasonable range, and I know the TR AI FTP correlates very well with my workouts and training structure. So I don’t pay much attention to the other estimates beyond the occasional curious glance, and focus my attention on my next workout and my overall progress.
Ramp tests overestimate my FTP because I’m on the wrong side of the 75% of your best 1 min bell curve or whatever equation each platform uses.
AIFTP was finally so frustrating to me that I hired a coach and left Trainerroad. I’ve trained ~550 hours this year and have finally gotten near the number AI FTP thought I was as a year ago, but now I have a huge foundation under it. Obviously not a solution for everyone, and the vast majority of people close to me get along with AIFTP just fine, so I’m probably an outlier.
As an aside, the intervals.icu model turned out to be the most accurate for me for FTP the whole time, but I never believed it could be that low because I drank the AI Kool-Aid.
Factually and definitionally incorrect.
FTP is the sole measure of your self-worth. Your entire entity and value on earth is measured by this number. The higher your FTP, the more your friends and family will love and respect you.