Bike VO2 Training: AI Told me to do this today

I was looking for some inspiration this week for indoor bike training and turned to AI Endurance.

My AI plan is set to be multisport-based but focussing on Ford Ride London (100-mile mass participation ‘race’) on May 26th. Interestingly the breakdown of the planned training has a lot of endurance work and the rest being Tempo and VO2max work. Like this…

That’s perhaps not a million miles away from what you might expect. Or maybe it is. Either way, the interesting take here is that the AI Endurance model has determined historically the kind of training I’ve done and the impact it had on my training metrics. The model then determines which stimuli work best for me and so those are the kinds it chooses for me.

You might say, ” Well they would say that, wouldn’t they?”. But AI Endurance also demonstrates how its model historically fits my data. As the following chart shows, it’s not perfect but the performance data and the fit data curves are usually very similar. This gives me confidence that AI Endurance is offering me something close to the best plan. Sure there could have been some magical kind of training I’ve never done, and the model couldn’t pick up and recommend it but my training is varied so that’s unlikely.

Anyway, this week’s training for both Tuesday and Thursday was 8x 2:00 @115%FTP with 2:00 rests, (might be a higher percentage as AI Endurance seems to give me a lower FTP/CP). Why that struck a chord is because Evan @NNOXX had told me that I need to do less Threshold work and more VO2-boosting intervals ie precisely like AI Endurance’s recommendations. Evan’s insights were based on some of my performances using the NNOXX SmO2/NO sensor.

NNOXX One analysed my half marathon effort – here is exactly what they found

I thought I’d tweak the session. The robots can’t always have things their own way 😉

I’d also recently used the Train.Red SmO2 sensor which is similar to NNOXX but with a neat feature on the app which dynamically tells you how to tailor your recovery durations between intervals depending on whether your goal is Stamina, Size or Strength. To cut a long story short, the Train.Red app determined on each interval that my required recovery varied around 45 seconds +/-5 seconds, and not AI’s catch-all 2-minute recommendation. Quite a difference.

Here are my SmO2 performance stats

The first of those two images shows that a good amount of time was spent in the R1 recovery zone hence helping stamina/endurance – exactly as intended. I wasn’t quite fully warmed up for the first interval but the second image shows my muscle oxygen reaching its lowest level during the 4th interval and that I was able to match that in the next 4 intervals. The peak of my SmO2 recovery got progressively less each time reflecting the compounding difficulty.

It’s worth noting that if I was aiming for a specific SmO2 recovery target I would almost certainly have waited for SmO2 to rise higher and hence to recover for longer and by the last interval maybe that would have been 2 or more minutes. so even with a novel and insightful muscle oxygen sensor, this shows that you need a clever metric (or app) to get the benefit.

[2024] Fourth Frontier X2 Review ❌6 Cons❌, ✔️8 Pros✔️ ECG heart rate monitor

I recently reviewed Fourth Frontier X2 which records continuous ECG during sport. It also determines heart stress from the ST segment of each beat. I don’t think I have a heart problem but I was sufficiently intrigued by the product to keep using it after the review period. Specifically, it can be used like a regular HRM but with extra ECG goodies packed in. I want to use it for more intense workouts like this one to be certain I’m not straining my heart, It didn’t buzz during the VO2max intervals so I know that my heart strain stayed within safe limits even though my absolute HR went reasonably high. Furthermore, the ECG trace only identified two ‘events’ that don’t seem significant. Here is the summary chart from Fourth Frontier.

Returning to the image from the top of this post, it’s from Garmin Connect. This overlays a more conventional power line that demonstrates that I executed each effort ‘correctly’. I did have a few more efforts in the tank but was fairly tired after eight, so I used the excuse of the AI recommendation of 8 reps to stop 🙂 Interestingly Garmin’s Stamina Line is shown in orange. It states that I was completely devoid of energy after 5/6 intervals and unable to continue…except I did. I do quite like the Garmin Stamina metric but only for longer rides and now I know why – it’s just wrong. It got this completely wrong, perhaps because Garmin serially failed to properly estimate my FTP. How difficult can that be? Even if it has properly estimated my Cp/FTP and I’m wrong then clearly it then fails to model stamina correctly.

I’ve also been looking at Garmin Daily Workout Suggestions and they have invariably been telling me to do base miles, base miles, base miles. It thinks I’m tired because it has my FTP too low! (I know I can manually set it)

Take Out

Training tech has come along in leaps and bounds over the last 10 years. Buzzwords like adaptive training morphed into AI training and I have to wonder what’s next.

I tend to use WAY too many gadgets, partly because of this site, partly because I get some enjoyment out of them, and partly because some of them even work to make me perform better!

My takeaway here is that some of these models are clearly too clever for their own good and might not be correct. Even with models like AI endurance, which I broadly trust, the workouts can still be tweaked with the right piece of tech.

Unless you have a reasonably good understanding of training to know when to challenge your digital coach, you might just be better off sticking to an old-fashioned, paper-base training plan

Today’s training? Base miles 😉