I’m curious how others are balancing AIE’s adaptive approach with the long-term visualization. Since AI Endurance is inherently non-linear and doesn’t follow a strict mesocycle structure, our synced calendars usually only look about a few days ahead.
I’d love to start a discussion on whether it’s useful (or even feasible) to have some alignment here. Even if AIE doesn’t use rigid blocks, it’s clearly targeting a specific load ramp leading up to our A-races.
Would there be value in having AIE populate the ATP with “placeholder” weekly loads? It wouldn’t need to set specific workouts months in advance, but just seeing a predicted trajectory for our Fitness and Form (TSB) would be massive for season planning.
Do you guys just “trust the process” and look at it one week at a time? I’d love to hear how the rest of you are handling this or if AIE has thoughts on bridging these two philosophies.
Actually something I’ve been wanting to implement for a while is that you could create week by week / (month by month?) guardrails to the training planner in Advanced settings that could be used optionally:
you could specify how much volume you want in Endurance, Tempo, Threshold, etc for a given time period leading up to the goal event. Then the plan finder would be constraint to work within those.
Although that’s technically counter to our approach to let the model find the optimal path, I understand that for psychological reasons alone you might want to block your training into certain (meso-)cycles.
That’s an interesting approach, though for me, I’d actually worry that setting manual guardrails might force me into a suboptimal path. The main reason I use AIE is to let the model find that optimal path for me.
What I’m really looking for is just visibility. I’d love to see the predicted progression of load/fitness/form over the next few months based on where the model thinks I’m headed. Even if it shifts as the plan adapts, just having that mental map of the trajectory leading up to my race would be nice.
Understood and totally makes sense, this post: Fitness tracking and curve goes in a similar direction. Will think of something that gets you more visibility while staying true to our approach
If implemented, I would like to see a side-by-side comparison between a cyclic plan vs. a pure-ML plan. My expectation is the pure-ML plan should “always” perform better than a plan with guardrails attached. If the mesocyclic plan is better than the ML plan, this implies something went wrong with the ML plan generation and it should try again.
My 2c: having an ‘advanced’ settings where guardrails could be implemented to allow for the creation of a cyclic annual plan (base/build/peak) would be for:
High-performance/coached athletes where athlete buy-in that they aren’t doing something so totally different from their competitors is important or,
Outliers who have a lot of Life Stress or do a lot of travel that is not currently able to be factored into plan generation by AIE. Being prescribed threshold/vo2max workouts during your two weeks of business trips is likely a bad idea even if the ML determines it is optimal at the time of plan generation.
Calendar planning improvements that allow for quick blocking of days/weeks/months with tags like ‘travel’, ‘vacation’, ‘year end’ to indicate high/low stress times of year
Athletes used to the base/build/peak method of season building. I would imagine such an athlete would prefer a tick-box that automatically applies such a constraint between ‘pure ML’ and ‘optimize within base/build/peak’.
Same kind of argument for why some competing products still use FTP to describe workouts even though their modeling no longer uses FTP for workout planning/athlete performance predictions.
