I have been using intervals.icu for a while now and probably just scratched the surface. I don’t know much, probably nothing, about AI supported coaching software, but am intrigued by it. If it works the way I hope they offer a training plan tailored to my abilities and ambitions, and automatically revise if needed. That’s my understanding anyway.
My question is, does interval.pro (or others) take age into proper account? I am 71 now and still take my training on Zwift seriously. But there is no denying that I need more recovery than I did just three-four year ago when I could race twice a week. Through last autumn I raced hard and successfully, got flattered by it and went for a sub-60 AdZ in December. Which I managed (by a 38 sec margin!) but was almost knocked out for the next month or so.
This week I worked out very hard on Tuesday and quite hard on Wednesday and Thursday. Today, Friday, I have a much needed rest day – but intervals.icu says I’m in the safe gray zone. Which could be true if I was half my age but my body certainly feels like I’ve been overreaching. Would intervals.pro turn up different results?
A coaching software that take my age into account would be good. One that doesn’t would just work me through the ground.
Partly a copy of another answer from me, but, her my 2ct
An LLM is a language model, not a world model. It generates probabilities, not truths. It understands neither your situation nor your goals. It only recognizes statistical patterns in texts and generates what has frequently occurred in similar contexts. It only generates plausible sentences. They also often reflect what you want to hear. It definitely reinforces your tendencies, even if they are destructive. An LLM can also tell you something wrong with absolute conviction without realizing it.
An LLM, on the other hand, can help you sort through your training data and notes and put them into a different textual form. It can summarize training logs or describe trends. If you don’t understand a concept related to training or nutrition, an LLM can explain it to you in simple language. It can also be motivating, thanks to its positive phrasing.
But it’s also often used to create workouts and plans. Can an LLM create workouts? Maybe. An entire training plan? Maybe, maybe not. An LLM does not know how to control progression. How much volume should it increase, how should intensity be distributed, periodization, regeneration, what do sensible workouts look like?
Mostly this is doner per rules „behind the scenes“. Strict limits must be set for LLMs so that they do not generate nonsensical workouts. If you let the LLM decide, I’m not sure if it will always come up with the best workout choice.
Regarding the fitness chart. It‘s just a model. No model and no LLM know if your feeling tired. If you feel like you need a rest, then rest! If you feel good for a hard workout, do it! Also look in the guide for the fitness chart explanation
Hi,
I conveyed your question to my project “AI director.” And here is the response:
“Your instinct is spot on — and worth naming: the “gray zone” is age-blind. It’s pure Form (CTL/ATL/TSB) math, which only knows your training load — not that you’re 71, not your HRV, not how you slept. So after three hard days the Form number can sit comfortably gray while your body is clearly overreached. That gap is real, not in your head.
That’s exactly what ICU Coach is built to close. How it uses age:
Recovery expectation scales with age. The model expects materially more recovery from the same stress as you get older — a rider in his 70s isn’t read like one in his 30s doing the identical sessions. This feeds the daily call directly.
It’s multi-signal, not just Form. It fuses load + HRV + resting HR + sleep + age, and can force a rest/recovery call when the signals say overreach — even when Form looks fine. That’s the part Form-only tools miss.
HR zones use your real max HR if you’ve set it (only falls back to an age estimate when none exists).
One honest note: it’s not a magic box that writes and rewrites a months-long plan. It’s an AI layer on your intervals.icu data that gives an age-aware daily read — “given your age and how the last days + recovery signals look, here’s what today should be” — plus plain-language analysis. You stay in charge of the season.
To your last line — a coach that ignores age “works you through the ground.” That’s the whole design principle: the same workload, read through a 71-year-old’s recovery curve, not a 35-year-old’s.”
As the developer of ICU Coach, I didn’t ignore the age factor because I am not young either.
You might be better off planning around a 10-day week, to allow more recovery. You already are aware, so it’s not something you deny and carry on blindly. Without knowing or seeing your data, it’s difficult to make decisions.
This was discussed on another (Sport Science) forum, where users used “AI” to generate a training plan. Based on the inputs it was given, it gave what each person had asked, mostly. Why?
Because they knew what they were looking for, and didn’t really need AI to tell them what they already knew. The real answers would be in what/how you ask it.
For someone who is totally new to training and wants advice, they probably wouldn’t be asking the right questions (prompts). They end with something, that they think is correct, but could be totally wrong.
The modern AI things based on GenAI, GPT and LLM are extremely good at giving the above expression, especially in the way they communicate with you. See, they are very good at working with words and human language (and even better with source code), but barely capable with numbers. So any AI based service worth its salt will use the AI part mostly for communicating, and do the actual planning the old fashioned way, i.e. with normal algorithms.
So yes, you can expect that the experience feels like communicating with a human coach. But no, there is no reason to expect particular magic results - and if those happen, then because the developer of the concrete AI tool you have been using has done a good job.
To be brutally honest, if you’re 71 you probably have time to spare between your training sessions. I would assume learning how training structuring works, and to learn to know your body signals etc., and using intervals.icu as data tool would be a more effective approach.
But if you want to see what all the hype is about, sure, there is an AI dedicated sub forum here, you might ask them directly if they support age specifically.
As @R2Tom has pointed out, you can show your form as a percentage or an absolute difference between fitness and fatigue. If you have a low CTL/fitness, and you’re feeling tired with it set to the absolute difference, then setting it to the percentage setting might be better. If say, your fitness is 30, then the absolute form setting will want you to keep your ATL/fatigue at 40 - 60, 50-100% more load than you’re used to, but if you have it set to percentage, the optimal zone will be in a more realistic, 33-39 range, 10 - 30% more than you’re used to.
If you have a fitness of >100, then the reverse is true.
Try swapping between the two settings and see if one shows you in the green to red, and the other in grey.
You could also just try making your own optimal zone for form which suits your age.
I have a custom chart that shows the %form in light grey and the smoothed/7day average %form and I aim to keep that line in the green zone. But the green zone could be set to whatever I choose. I find the default 10-30% is sustainable.
You can see I let the line get into the blue before a big walk a month ago then took a week off for the line to get back into the green and then, whoops!, didn’t quite get back into exercise as soon as I should have and then whoops! again, got sick, but in general keeping the line in the green is sustainable for 60yo me.
My age is 74 and because of that I use a AI tool to train save. I am using after some other trainingtools the AI coach Koda coach. After using it for some months I think it’s great. Highly recommended
I’ve seen this someplace else, %age vs absolute numbers, but that was a great explainer for a somewhat noob on limited hours. Thanks. Got the intervals and strava MCP running and I’ve stopped manually prescribing myself workouts. Just tell it how I feel and it tweaks my diy set up plan every week. Don’t actually need the strava MCP except for historic data but that’s not very useful.
This is very far from my experience with frontier LLM models like OpenAI GPT-5.5 and Claude Opus 4.5+. Have you ever presented 9 wrong progressions and 1 right to such LLM models and asked it to rate them by correctness and explain the reasons for their scores given an athlete context you describe? (age, objective etc?). If you do so you will quickly see how well these models can judge plans and explain their reasoning.
I won’t do this, because I know what’s an LLM and how it works.
But for you, I’ve just presented a 4 week plan with a completely wrong progression of vo2max workouts, and additional 9h endurance ride, and on top a 3h “endurance ride” at 90% FTP. The only thing it would change is doing this 3h “endurance” ride only every two weeks. No recovery at all. The progression was totally bullshit, but didn’t noticed it. It just confirmed my “wrong” view of the plan.
At the end, there doesn’t exist any magic plans or workouts. Why should I ever ask or create it by AI?
Thanks for the great response, everyone. Now I have a much better insight in how these models work, and some of your comments have enabled me doing a lot more precise search.
I have also got better confidence in just using my own training experience and trusting my subjective perception on how my body feels. I cannot remember a life without working out, and AI may have a hard time coming up with better info than what I already possess. I may have had unrealistic expectations.
What I certainly will do, is changing the fitness settings to percentage rather than absolute numbers. So a special thanks to you, Samantha.
