I used PKRS for my first race of the year. Paid for the coach version. Did pretty well at the race. It really bumps down your workouts based on how you're feeling. It will strip the intensity or cancel workouts if you report that you're tired. It seemed really polarized, mostly Z2 and Z5.

there was this thread recently

---

Brian Stover USAT LII
Accelerate3 Coaching
Insta

Thanks, I'm interested to see if anyone has tried the alternatives also.

These automated training programs seem like a no brainier. After all most coaches would have a similar generic type outline, with the variables being heart rate, power, speed, frequency and general athlete feedback of fatigue etc.

On a Sunday what are the things that a coach looks at to determine what the plan for the week is, shouldn't be too hard to automate this aspect?

The problem is that none of these applications have the data necessary to do anything interesting/differentiated/better compared to a coach or even just some program you buy off TraingPeaks.

---

Dimond Bikes Superfan

Good coaches use more than just the data to make decisions. The data might show X, but you might be feeling Y. Also, every athlete has different habits, traits, etc. For example, I worked with a coach for a while who eventually discovered thatanytime we crossed over a -30 TSB, I'd get sick. Would an AI platform show that? Possibly. But having human discretion is still a valuable part of the equation, IMO.

---

@floathammerholdon | @partners_in_tri

The way you’re feeling is still data and at some point there will be a platform that is capable of taking that into consideration. There’s plenty of blind spots in AI generated training plans but your example is not one of them.

But the plans off Training Peaks aren't adaptive are they? Where as I know with Traineroad it is constantly changing my plan based on what Im doing. I assume the others like Athletica AI etc are also.

As for coaches from what I can gather the overwhelming majority don't offer a regular one on one weekly consultation without a premium price. Even then what are they getting from the convo other than a gauge for fatigue?

Nothing against coaches and I trust there is some nuance that automation can't take into account, but for the overwhelming majority how do they plan for their athletes? They wouldnt be reinventing the wheel every week, they would have a general structure and then modify it slightly based on if the athlete is completing all the sessions and or their hr and pace etc-all aspects that can surely ( and have been...?) automated?

All these AI programs ask how you are feeling and adjust for reported levels of fatigue.

It would be interesting though if it could pick up trends like your example where every time you crossed over -30 you missed sessions and therefore automatically took this into account moving forward. Again if it doesn't currently Im sure its something that they will be able to in the not to distant future.

Definitely interested in people's experiences with these various programs. I'm happy with my 80/20 plan that I use via training peaks, but I think starting next year I'd be inclined to use one of these AI run programs as I progress in my fitness.

an AI can take into account anything that can be measured, even if only subjectively, which is to say pretty much everything, it just depends how well programmed it is in much the same way that a really good coach will collect and incorporate lots of things into the planning, while a lesser (or cheaper) coaching plan will be more of a standard routine. the question is whether a given coaching service, be it AI or human, does in fact take into account all the relevant metrics. there is a bit of a difference in that AI at some level has to be programmed to understand relationships between factors such as TSB and compliance, whereas a human with true inteligence can work these out upon seeing them

another factor that many coaching services of either sort do not really take into account is the specific demands of a particular goal event and how that compares to the athletes strengths and weaknesses.

where i find AI tends to struggle most is with scheduling around complicated availability constraints and the overheads of training sessions eg travel to pool, showering etc

AI can’t draw faces or fingers very well either, so when it’s looking at you, it usually thinks your glocose/maltodextrin ratio is off and that you could probably use more reach in your position.

After a couple of seasons using the 80/20 plans from the book for 1 year of sprints and 1 year of olympics I'm gonna give Athletica a go for Oceanside 70.3 next year. I'll let you know how it goes.

Before adaptive training, I found TR on the verge of being too difficult sometimes but made big strides. With adaptive training and honest answers at each "how did that feel" and "why," I find TR perfect. For the cost, I don't see much benefit to getting a coach. The only coach I'd want is a swim coach because that isn't purely an effort-based sport.

Yep I agree trainerroad had been great, but feel it neglects the run and swim aspects. It provides constant feedback on your cycling progress, but run and swim are nothing more than neglected poor red headed cousins..

So if there is a platform that is what TR is for cycling but across all three disciplines well shut up and take my money!!

While with adaptive training you’re less likely to overdo it (excessive volume of intensity is something that TR used to be “famous” for among triathletes), the general training prescription where they insist on having a number of moderate to hard workouts per week (on the bike alone, plus running and swimming) goes against what I’ve heard from experts.

Two half-assed, undercooked “threshold” or “VO2max” sessions in a week (undercooked because adaptive training dialed them down) are not what I’m looking for.

What I’m looking for is one solid session.

As a matter of fact, the biggest gains I made the year that I left TR and implemented a much easier cycling plan. The gains were in running off the bike as well as cycling.

---

"FTP is a bit 2015, don't you think?" - Gustav Iden

How long ago was this?

I guess the tricky thing here is that indoor riding with a power meter is such an isolated/controlled thing making it easy to adjust etc. Power is just such a good tool and does not work in the same way for running.

What does a coach look at when evaluating run performance, speed for a given heart rate, tracked over time? Ai should be able to evaluate the same in a more accurate manner?

Must’ve been 3 or 4 years… I stayed subscribed after that, for a while, but was only doing custom workouts.

---

"FTP is a bit 2015, don't you think?" - Gustav Iden

Yeah your feedback seems very similar to many from back then. I think you will find it's heaps better now, the adaptive training is great.

I was still subscribed when adaptive training was already around. Just wasn't necessarily using TR's plans. I'm sure they're better with adaptive training (as they were then) because you're less likely to overreach or overtrain. The "AI FTP" estimate that TR spit out made a lot of sense to me. Group workouts were a terrific feature - on the rare occasion that I had someone to train with. Still, for the reasons stated in my earlier post I only used TR as a trainer app, and a year ago I switched to a different trainer app.

---

"FTP is a bit 2015, don't you think?" - Gustav Iden

I've been using Athletica AI for a few months now in prep for my first triathlon next year, a 70.3, as someone who's never run more than 6 miles and who hasn't swam in decades. Cycling is my main sport.

It took me a few tweaks to get to a plan that seemed right to me, right now I'm in a season where I don't have quite as much time to train, and my A race is a little under a year off, so I'm in a low volume program for triathlon. With the low volume plan there's no scheduled strength training, so I'm manually going in and removing swim sessions until the spring and adding in strength (for a number of reasons).

So far I've found it's doing a great job of updating my sessions to match what I've already done. If I go out for a MTB ride instead of a HIIT cycling workout and it runs long, overnight it'll update the sessions for the rest of the week to match.

Also, it won't move around rest days as needed, but it's pretty easy to move things on the schedule to match yours.

I've worked with a good number of coaches in the past for bike racing, so I have more knowledge than a beginner, and understand the whys behind the workouts, what I was really looking for was an affordable, adaptable program that gives me guardrails for workouts each week.

My trial is up pretty soon and I'll be buying a subscription for sure.

A great point. Trainerroad likely has the biggest data set of any of them to base it's algorithms off and it's still hit and miss. Much better for single sport cycling, but multisport seems like an afterthought and I don't think TR have any intention to change this. Great workout library and calendar though, and the Autodetect FTP is good and comparable to a ramp test (usual caveats that it's likely too high for most people).

I agree and just don't see how these smaller apps will cope without a significant amount of manipulation from the developers. I may give athletica a go to see how it goes.

The trouble I have is that all the "AI" stuff is proprietary, and not vetted by independent research. E.g. with the Banister/Coggan stuff we know the analytics behind it, and we can point to a wide body of research.

With "AI" all we have is anecdotes from people who try it. We don't even know it's really "AI," e.g. there are numerous ways to make "adaptive" algorithms using non-AI models. Wahoo SYSTM does this explicitly.

I understand why all the companies do it. They all saw everyone "steal" the non-proprietary Banister Coggan models, and those guys get little but academic accolades. Even Coggan himself went proprietary in the more adaptive training-data-driven stuff in the newer versions of WKO+ (I think he's the one beyind that stuff).

I could blame academia for falling behind the state of the art, but it's hard to blame them for not having access to data. This stuff requires tons of data - can't just do the typical thing of recruiting 25 undergrads to do a few time-to-exhaustion tests over 3 months.

So we're stuck with "TrainerRoad is amazeballs!" Which isn't quite the same as RCT trials.