Scenario.
Your workout for tomorrow and look for longer or harder options in the options. When you select a longer workout (15min longer), your anticipated FTP at the end of the block is unchanged.
Do you leave the shorter workout as a minimum effective dose? Why do more work if the net is the same….
Or do you increase the dose for the other intangibles, felt good, had time, working to longer duration event, etc.
Does this question even make sense? Do you subscribe to this MED principle?
If the prediction stays the same, I will usually opt for the longer workout. If the prediction drops, I will usually go back to the original workout, but sometimes, even with the original workout, the lower prediction stays, so 
I think “minimum effective dose” is a flawed approach for training in some (many?) situations. I think a better concept is “law of diminishing returns”. More training is generally going to result in more fitness (if that training is well structured, proper rest, etc.). As we add more and more training stress, additional fitness comes slower and slower (but can still increase). At some point, more training will become counter-productive and results in a fitness/performance reduction. But improvements can be made up to that point and most amateurs are nowhere near that point (IMO). Minimum effective dose is a very conservative approach to training that will certainly improve your fitness without pushing into the risk of excess fatigue, overtraining, etc.. But it’s not an approach that is going to maximize fitness/performance. If you look at ramp rate in your training, that’s a key metric showing how quick you are increasing training stress. As long as you are ramping stress at a slow rate (ie - minimum effective dose), that should translate to fitness improvements. But if you are ramping slower than you could, you will not reach your potential peak before your event/season is done. The trick is finding the maximum ramp rate you can consistently recover from. TR is very conservative on ramp rate, which favors slower results with lower risk of burn out or over training. Not a bad approach, just not a fitness maximizing approach (IMO).
I think i agree and really this is irrespective of the whole AI thing. But bc of the modeling capabilities we now have, we can see, in theory the result of various approaches.
I would not change the workouts AI has assigned, adjusting the duration will have negative effects on your fitness long term
The key being “in theory”. AI does a nice job of pickup up patterns in huge complex data sets. And I don’t doubt that the TR system can help figure out the right amount of training stress for an individual based on pattern matching and trying to get that athlete into the right bucket. But the TR system is also working with very limited inputs around power/time/hr/etc. Not saying it won’t get to the point that it can actually predict future performance based on different training scenarios, I just think it’s a bit optimistic/naive to think that is totally baked in an initial release from a small company with limited investment. That’s not knock on TR, I think it’s great innovation story and the approach/concept is great, I just think it’s in it’s infancy and will take some time to mature.
If any change in duration was bad, then there wouldn’t be a workout alternate option, and it would be impossible to change your volume ever. Clearly if the AI is telling you to do a 30 minute ride, the 2 hour ride isn’t the optimal choice. But adding 15-30 minutes to a workout if you have more time that day, is a good way to build durability and increase volume.
Kind of but not really, maybe. 
On one hand, I’m generally following a TR plan more than ever and not throwing in additional hours, as per previous years.
On the other, I do use the new tools and the prediction to guide me when I find myself time crunched. If the planned 90 minute ride is going out of the window, will 45/60/75 minutes retain the same prediction? I’ll still try and ride to the longest option that my time allows (without dropping the prediction) but every now and then, I’ll choose the MED.
The volume should increase throughout training as the AI increases it, not randomly though. Adding 30m to workouts is just extra fatigue that impacts future training sessions.
The suggested durations are mostly a function of how long you told the system you wanted to train each day. It’s not because they are the durations that drive the best performance. While you might have negative effects from significantly ramping duration too fast, increasing your duration (and overall training stress) is going to result in higher fitness in almost every situation when done properly.
I think the AI will stick to your schedule that you built out when you started your plan, unless you run the “check volume” option. The only workout length that will change automatically is the new “dynamic endurance’ option. The AI will suggest longer workouts, but only if you have maxed out the progression at that length of time that you have in plan builder.
I’m finding that 1 hour sweet spots get pretty routinely increased. But I’ve not seen it for any other. I think it’s because if you don’t increase the duration it’s just ride 50 minutes straight at sweet spot. 
So FTP and by proxy predicted AIFTP is not prescriptive. It’s a useful guide to stop you doing stupid things. All those people that insist everyone should do 15hrs every week all year because that’s what they’ve always done, for example.
“The right workout” is a mixture of the work you did before, the work you will do, the recovery and nutrition you’ve had…and the time you have available and so on.
So no, don’t make decisions on workouts purely on the consequent change in predicted FTP.
Yes, do the minimum work you need to stimulate adaptation, within the principles of progressive overload and your readiness on the day. Humans are super compensators. TRAI is trying to do that calculation for you.
I’m in the “do what it says for max improvement for your event“ group but I’m doing like a 12 hour thing and just can’t help but push out some of my workouts with “longer” alternatives. At least I’m sticking to the recommended optons and not going totally nuts. Usually.
Joe
Doing more volume (assuming you are keeping the intensity to a level that it does not negatively affect your key hard sessions) will be beneficial.
Measuring improvement off of a single number will not give you enough detail to know if your training is improving your fitness or not, there is a lot more to racing than just FTP.
Totally agree, but the key limiter you have there is “the time you have available”. If you increase the time available, almost all athletes will get stronger with more training volume if it’s done properly. There is zero debate that training volume is one of the biggest drivers of developing aerobic fitness/performance. A time-crunched approach is the right approach when bike racing isn’t a career and/or we have other priorities, but nobody is claiming that a time crunched approach is a path to peak performance (unless then are selling something).
Maybe, maybe not.
Maximising dose is a fools errand.
The data is pretty clear on more volume being a huge driver of performance gains. It’s an aerobic sport after all
This is why I think showing the FTP prediction does more harm than good. You start playing a game to bump the number up instead of just training and following the plan.
If you are talking about maximizing does in the context of overall training volume, the vast majority of us amateurs are so far away from approaching that limit that it’s a bit of a silly argument. But yeah, I agree that maximizing overall training volume is a risky/dangerous approach that even the pros struggle to get right.
I think the place minimum dose is a more sensible approach is at the workout level when focused on shorter term acute adaptations. Adding on extra minutes/watts to a v02max workout every time is often creating more acute fatigue cost vs. the benefits/adaptations it might drive. I’d still say the “minimum” isn’t a great approach, but that’s why tools like TR give you options to be more or less aggressive in your training. Minimum slowly builds fitness, but can leave gains on the table in the limited time you have. As you get more aggressive, there is decent chance you will end up with better fitness, but with less room for error/hiccups and more risk of failure. Training is like anything else, you have to take risks to approach limits. AI isn’t going to change that anytime soon.