Cycling training AI madness

TR Threshold level 3 (60 minute) workouts range from

62 TSS / 0.79 IF (10 min steps @ 60%, 70%, 80%, 90 and 100%) with 30 second recoveries between each step

to

85 TSS / 0.92 IF (3 sets of short efforts lasting between 1-4 minutes at 100-115% FTP with very short, 30-second recoveries between intervals. Recoveries between sets of intervals last 2 minutes each.)

Frissell -2 is in that list, 74 TSS / 0.86 IF (4x9 at 95% with 3 minute RBI)

When deciding if you could complete such workouts, keep in mind that unless you know what your TRAIFTP is, you are deciding blind.

For context, consider that 2x20 workout that you posted here the other day

The TR equivalent of that could be a workout called Gray: “Gray is 2x20-minute intervals at FTP with 5 minutes of recovery between intervals.”

That is a TR Threshold level 5.9 workout. 84 TSS / 0.92 IF

FTP is quite the can of worms for training platforms. I have my opinions but people like high FTPs. Live and let live I guess.

I’m a natural sand bagger if anything. Doing killer workouts due to inaccurate FTP is something I experienced and I hated it personally. It has been quite the rabbit hole of learning since then.

If I recall the pithy power proverb correctly we should train WITH power not BY power.

Dave

1 Like

I agree.

I would rather miss out on a small extra gain than go over and end up needing a prolonged recovery period, or, worse, burning out.

I think the TR crowd likes a certain feeling from their workouts. The workouts feel effective, logically makes sense to the audience. And my guess is that’s how they base things. That’s not a terrible approach, especially if they are no longer burning people out. I would say starting at level 3 in December and progressing, you likely peak by March or something. So it depends on when you start and what your goals are. But again most people joining want to feel they are doing effective workouts.

This is all very experimental (to be truthful most training advice is). From my understanding, they’re still fairly heavily sweet spot based. That’s kind of been scientifically “proven” less effective than some other methods. But I would say it’s also validated by many of their users that they’re happy with the results.

I think most TR users feel that they’re time crunched and they should be doing something different than elites, and higher TSS contributes to that solution.

So I believe TR has improved a lot over the last 4 years and I would recommend it a lot more now that I would have in the past. I don’t like this AIFTP concept but if you are 100% bought into TR it’s likely fine. I am not so in love with strict FTP either, or such a heavy focus, maybe this lightens the rigidity which is good.

1 Like

Here’s a funny complaint

https://www.reddit.com/r/trainerroad/comments/1qlnwvi/trainerroad_ruined_by_vanity_numbers_and_gimmicky/

One other new feature is “Dynamic Endurance.”

There has long been a lot of noise from users that their training plan did not include long enough rides to prepare them for the events they were training for.

Now users can set the upper time limit (5 hours is the maximum on the drop down list) for their endurance rides and TR AI will decide how long / hard a ride to prescribe. The intention being to gradually increase the duration throughout the plan towards the maximum.

The AIFTP prediction feature is easy enough to turn off. Complaint fixed!

Yeah, I see that later in the thread. People don’t even know it’s possible to turn it off, I think is the problem.

1 Like

Hopefully the TD AI program can meet the needs of those who don’t want either CJ or self coaching.

Dave

1 Like

People will always have ideas of what they want. We won’t fill all the gaps. In general, we just try to keep things low cost and do as much as we can within our constraints and that we feel makes sense.

1 Like

The AI cycling training perception hasn’t really gotten any better since January.

There is an interesting poll on the TR forum, roughly split 50/50 between AI made the platform better or AI made it worse.

Dave

Except where it falls down is to train AI on the subject you have to be a subject matter expert.

They sacked their only ‘real’ coach with training knowledge, so it is trained on the CEO and his staffs beliefs, not on science related to performance outcomes, personal beleifs and what would sell to outcomes.

I was going the write a lot of examples, of beliefs, bias and outcome, but you just need to watch a few podcasts to get it, and make your own mind up.

Yes, I greatly appreciated what Coach Chad taught, he was a bit too Coggan focused but you could see him evolving even while TR was still stuck in high volume sweet spot. I believe if TR just evolved as Chad was they would have been even better.

1 Like

Diplomatic.

I think Chad saw what was coming and maybe didn’t do what he should have as an employee but that is pure speculation.

It appears they are really suffering. It seems like much of their own forum traffic, Reddit, etc… is dissatisfied. I have the impression they have zero interest in moving away from AI, but the market has spoken and it doesn’t want it. At least not the way they’ve implemented it… Even if they did want to leave AI they are so dug in at this point it doesn’t give much room to manuever.

Endurance training IMHO is not that complicated unless you’re trying to be at the pointy end and optimizing yourself. And if you need that level of optimization you need a coach not an app.

I am a lurker over there at times just to read anything interesting about training, but the content has taken a dive.

Dave

Yes I am watching too, and your 50% survey results is interesting although the complainers are more likely to respond is my guess still it’s bad and they kind of painted themselves in a corner… it will be interesting if they can figure a clean way out.

I wonder if the proportion of users complaining about TR AI is larger / smaller than the proportion of users that complained about the intensity of the sweet spot plans back when they were the thing?

Coming to TD from TR, I’ll say I did not like their AI plans. That’s not a dig on the tech, but the execution. The expectation was not meet, that being the plan ought adapt to individual metrics with an objective. From my sample of 1, the AI plan did much worse adaption than merely selecting from one of the three AI Train Now options. I’m not the target audience for TR, or most cycling apps I gather, fitness for fun and health.

With respect to AI, my preference is BYOAI or at least have that as an integrated and supported option. The tech is moving too fast to pin down a solution, better to allow for more experimentation. I’d be surprised if devs of fitness apps are up to date here, running a Hermes, Pi, OpenClaw, or other agent 100%, knowledgeable of their architecture and capabilities.

Yes, I think BYOAI is the way forward.