By accident Youtube served me a snippet of this video on a new AI-enabled cycling training platform called Vekta. They claim to be working with pro teams and it seems like a mix of TR and TP, but with more advanced analytics tools. They make some big claims, but judging by the screenshots, the app/platform looks very capable.
The design looks clean, it starts from the assumption that you might have a human coach and tries to put relevant performance metrics front-and-center (and both are different for coaches and athletes).
Interestingly, they have publicly shared their roadmap and seem to be plugging away. Members can apparently leave feedback on the roadmap.
Overall, it seems like something where TR wants to be in a few years’ time (judging by what @Nate_Pearson has said publicly on the podcast) … assuming it works.
No idea. One of the touted features was automatic interval detection, so it seems to be outdoor-first. I’m tempted to give it a try, but wanted to see if anyone has heard of it.
They have some cool integration ideas and the UI looks very nice. Digging through their website, it does have a workout creator, but it doesn’t appear to me to have a plan builder or a workout player, and the post-ride AI analysis appears to be similar to Strava or FasCat in that it just tells you what you already know. Cycling only for now, but running and swimming planned for the future.
I might test it out for a month on the free trial because I’m curious, but with no training plans, it would be a non-starter for me unless I just absolutely hated using GarminConnect/TrainingPeaks/Intervals.icu.
They also have nutrition on the planned feature list. After seeing parts of the interview, this seems like a very serious attempt.
If you ever tried it, would you be willing to share your feedback? I might give it a try during my next off season/shortly after. But it isn’t quite clear to me
One of their features is called AI session planner, and I assume session = workout?
Edit:I see what they mean, they used an LLM and standard ML APIs to generate workouts from a description.
If these metrics were presented in a more user-friendly and more easily trackable fashion, this would be a net positive. intervals.icu is extremely powerful, but you need to know what to look for.
Yep. Agreed on all. So no plan, but if you say “give me 4x4 VO2 with 5 minutes rest and 10 minute warmup and cool down”, it will build a workout. Fascat does that too, I think. But, if you’re going to do all that, I think it’s faster to just build or edit a workout in TP/Intervals.
I find these “HEY WE HAVE AI” when it’s just a backend for meaningless garble fed out of ChatGPT interpreting your workout to be totally non-useful features.
However, I do like the integration of the ML in AI Endurance. You can go in and ask it things like “Why have you used x workout to define my LT2, can you remodel it based upon this other workout”, or “What might be the best set of intervals if this represents the selection at my race”, or even “I’m feeling unmotivated today, can you reshuffle my plan so that I have lower intensity for a few days”.
So the AI can take over some of the busy work of coaching (creation of standard workouts, do standard adaptations of the training plan) and reason about why it did what (within some limits). That’s neat. So it seems as if TP should be worried more than TR.
Nate mentioned that he hopes TR will be able to do this at some point. However, TR will very likely need major re-engineering to introduce concepts like coaches and make it coach-centric (to introduce coaches and coached athletes as well as the amenities you expect these days).
If this were a start-up with zero customers (like, say, replacement-for-Zwift-with-no-active-users), I’d be more skeptical. But it seems they are serious people working with other serious people, and if the screenshots correspond to the actual apps, then I think this could be an interesting addition to the market. TP’s UI is ancient and not fun to use.
Sorry if I wasn’t clear, I was referring to AI Endurance - which has a sort of internal LLM that can also do a lot of things with the AI as well, like say “I’m feeling a niggle in my calf today so can you swap my run out for a swim that hits the same training effect in x hours” and it will work it through.
Basically it can read and understand your plan and make sensible recommendations. Obviously, like many LLMs, these can end up being a bit silly sometimes, but mostly I think for a lot of people this kind of functionality is all they really want or need from a coach - someone or something to give a level of sense checking to what they are doing.
I tried the fastcat one and it’s really limited to the kind of Strava vague positive vibes summary.
So just to confirm (as I haven’t jumped in yet to look), this app doesn’t have the ability to create any training plans?
The breakdown of CP W’ Max Power seems very similar to Xerts signature right now which is what I’m using. This looks to be a much cleaner interface vs Xert but at least it can create me a plan to follow.
Just took a look. Doesn’t show a workout from Zwift yesterday. Doesn’t offer any plans etc. Maybe it’s useful if you’re a coach, it’s utterly useless to me.
This is really a TrainingPeaks replacement as far as I can tell. It is already being used by some World Tour teams.
They have replaced FTP and TSS with their own metrics, which they claim are better and more accurate. Access is free for coaches and the athlete pays a subscription. There is no plan builder as in Trainerroad - you can get it to build workouts for you but you need a Zwift (or alternative) subscription to do the workouts. There is a podcast with one of the creators for a better explanation: https://www.youtube.com/watch?v=I_QFSnTga7c&ab_channel=TheRoadmanPodcast
I heard about it on podcasts and elsewhere. It’s pretty sub-average imo. It’s claims of world tour are imo potentially similar to SIS and stages was to team sky. I tried hard to make it work. But free strava + intervals + copilot makes it look pretty silly, pretty fast. Uninstalled and do not recommend.