For the load ramping: actually how you describe it is not how we have designed/imagined it. It should do exactly what you where hoping for and slowly increase your volume and your longest activity duration depending on your last 6 weeks of data in the system. Could you please send me the email address you used to create your AI Endurance account to markus@aiendurance.com? I really want to have a closer look at your account to see if there’s a bug or something is misconfigured for your case.
I like it and plan on using it for this training block, but the automatically detected thresholds are way off. It autocalcs my threshold and v02 max paces slower than what I can and have comfortably run 2+ hours at. This happens whether I use HRV derived thresholds or not. By default it had my v02 max intervals at around my regular easy pace.
I can set my own thresholds but it brings the intelligence of the platform into question if it thinks I’m going to get a big training stimulus from 5 intervals of 1 minute at easy pace.
How does AI Endurance take account other sports like cross country skiing, skating etc. ?
As triathlete living in Finland I am plannint to do big part of my training XC skiing next months.
Thanks - We do what we can in the autodetect with the past data although there’s some improvements planned in the near future. If using/not-using HRV thresholds doesn’t make a difference there’s likely no HRV thresholds detected yet. Would recommend using manually set zones for now in this case.
Everything you track we do take into account for the recovery model via the training stress (ESS). Would be nice to use the time doing other sports in our prediction/optimization models too but we’re not yet doing that
Hi @Markus_Rummel_AIE
I see differences in target power between AEI and the corresponding workout in intervals.icu
Generated a new plan in AIE.
Tomorrows ride shows a target of 163W
After trying to use this platform for both running and cycling training I decided not to renew the subscription. I found myself constantly fiddling with knobs and adjusting my threshold pace/power trying to get it to build a training plan that made sense. Often it would generate a plan that predicted no performance benefit from training for months until the last couple weeks (presumably due to taper).
I never felt like the model accurately assessed my current performance or the training effect I would receive from any given session. I think the product has good potential to be a valuable tool similar to TrainerRoad for runners and multisport athletes but the model needs a lot of refinement.
It also feels very closed off and static. I don’t think there were any updates to the website the whole time I was using it. There’s no news or announcements when you arrive on the website, there wasn’t a single new blog post while I was using it. There’s no internal forum for users to discuss the product or ask questions, there’s no public roadmap to show what updates, if any, are planned.
Sorry it felt like too much fiddling, we’re always trying to make it better so thanks for the feedback.
Apart from our newsletter there’s a Reddit and a Facebook group you can join where we answer a lot of questions and we also share a lot about what we’re currently building. The links to those are also on the homepage and should have been send to you via email after you signed up.
We’re working on a couple of bigger features now:
an agentic chatbot that lets you make modifications, gain insights and perform deeper analysis. Think our GPT with also write access as the goal.
measuring and optimizing durability with power/pace vs HR/DFA a1/respiration rate. This is also an active research project as we mentioned in the last email.
In general we try not to spam you with ‘marketing stuff’ like many other apps but focus our energy on actually improving the app and coming up with completely new features/ways to leverage your data that in the end benefits you more, or at least so we hope
No problem. There’s a couple of things you could try:
Exclude data that is inaccurate (for example off power meter or not representative of your fitness) via the Data page.
(De-)activate “Use secondary layer of training for digital twin” in account page. Using the secondary layer is in general a good thing (more data) but can sometimes lead to higher error estimations if you’re off what’s typical.
Experiment with using pace or running power: running power can be inconsistent between recording devices (for example Garmin running power is way higher than Stryd for the same effort) which can lead to large prediction errors.
Hi there, I’ve been an active subscriber for the last few months.
My main complaint is structure swimming workout not properly parsing in interval.icu: the structured workout syntax is way off and I have to manually edit to make it works.
Currently the workout are copied in this format (not recognized by interval as a structured workout!):
Hm that is odd formatting indeed. And it’s formatted correctly in AIE? Just checked and we are exporting the same way to Intervals as we would display the description in our app.
If I remember correctly, when we did the integration a few months ago Intervals did not have the option to export a swim workout in a structured way, just the workout text. Is that still the case @david ?
Interesting concept. I have signed up for AI Endurance and tried to read up on how to use it both here and other sources. I have connected intervals.icu, Garmin Download and Strava. I have also installed the AlphaHRV app on my Garmin (and hopefully set that up correctly). I should also say that I’m primarily focusing on running and I run with Stryd power.
Now, these are the questions I have:
During winter I have done a lot of cross-training, e.g. strength and cross-country skiing. As a result, I’ve only done 2 runs/week, but overall aerobic fitness should be largely maintained. However, there seems to be a ramp-up phase with lower-volume and lower-intensity that doesn’t match my current fitness. Can I do something about this?
Also, the zones were set way too low when detected automatically. When, I set them manually I get a notification that my plan should be recalculated, but this seems to trigger an auto-setting of the zones and then I’m back to start. I have chosen to decline this recalculation.
I have updated previous CP tests with “test” and duration in the title. However, I have chosen to do some of my tests as 5km tests rather than 20 min tests. Can I indicate 5km instead of duration? Is there any way to re-trigger a model update after updating these titles?
I’m surprised that there the data for the last 2-3 months has weight=10 even though I have not had AlphaHRV installed. Is this still valid data? Will AlphaHRV provide better data and hence improve the model? Is there a benefit of saving the AlphaHRV Readiness measurement to the fit-file?
What happens if I decide to execute another workout than the one suggested by AI Endurance? Should I then mark the workout as “skipped” in the calendar? Will AI Endurance then pick up the workout I actually performed and update the plan accordingly?
For executing run workouts it’s actually useful to have workouts structured with a power range (i.e. 250-270w) instead of only one power (e.g. 260w). I understand that this is different than on an ERG trainer. Is there a way to get such ranges for running workouts?
The fitness-chart on Intervals.icu requires an FTP value to be set to calculate fitness/form/etc. How should this be updated when using AI endurance?
Thank you so much for taking the time to respond. Looks like you’re onto something very interesting here.
No I don’t think that can be hacked - certainly shouldn’t - even if you have fitness from other activities it’s probably a good idea to ramp up especially the running slowly.
Yes best to handle like you did. In rare cases this can happen.
For CP it’s actually not necessary to label it as test, it should pick up on it. No direct way to trigger a model training but almost all updates that should trigger a model retraining automatically do.
10 is whenever it’s a good measurement. For example, for CP this could be if there’s at least one CP relevant effort in the 2 week window where HR was a) high quality and b) high enough to be detected as a proper all out effort. We calculate readiness ourselves from the raw RR interval data so no need to save it to the fit file unless you also want to see it in Garmin Connect.
yes exactly, it always adapts to what you actually did and in that sense does not expect you do the training as prescribed
If you export the workouts to Garmin (connect Garmin Upload in Apps page) the workout execution is with power ranges directly on the watch. All our workouts are with ranges, Endurance workouts with “keep power less than x”, Anaerobic intervals with “keep power more than y”
If you’re using Intervals.icu to view the workouts exported from AI Endurance or to export to another platform we don’t connect to you should set the Intervals FTP to the one we advice in Account page, Zone section so the power values are consistent. If not you can keep Intervals FTP independent of AIE.
