I can’t find the original comment from sloan58 that you’ve quoted and would like to read it in its full and original entirety. Was it moved elsewhere, did the author delete it, or are forum posts critical of the new system being deleted? Can’t seem to find it.
Just wanted to say I am very psyched about this update. I’ve subscribed to TR off and on since 2020, also had IRL coaches off and on since 2021, and had bouts of self-coaching. I have always fallen into a pattern of trying to do everything a coach gives me and saying (and thinking) I’m fine until I suddenly and completely implode and stop riding altogether for months. When I self-coach I tend to take things too easy. I’m excited to be using the new TR model. hoping it helps me dial in the right training load. Thanks for all of your hard work on this!
He deleted his own posts. What he wanted was added to the list of things to fix.
Not reading through hundreds of posts, so I used Gemini to summarise the changes, particularly addressing some of my concerns. It helped me, might help others.
TrainerRoad AI Update (Jan 2026) - Comprehensive Summary
The latest major update, branded simply as “TrainerRoad AI,” was released broadly around January 21–22, 2026. This is a complete rebuild of the training engine, moving from simple logic trees to a neural network trained on millions of rides to simulate how training affects your unique physiology.
1. Key Changes in the Release
• Simulation & Prediction Engine: The AI runs hundreds of simulations to predict how a workout will affect you in the future. It can forecast your fitness (FTP) and fatigue up to 28 days out.
• Raw Data Analysis: The AI now looks at raw power and heart rate data directly rather than relying solely on “Workout Levels” or “Zones” to understand your fitness.
• Predictive Fatigue Detection: It forecasts “red days” (high fatigue) before they happen. If you do a hard unplanned ride, it recalculates your future fatigue and may suggest a rest day to prevent burnout before you crack.
• Dynamic Endurance: You can set time caps (e.g., “I only have 2 hours”) and the AI will scale the difficulty and duration of your endurance rides to fit while ensuring progression.
2. Benefits for the User
• Reduced “Failed” Workouts: The new engine is 70% more likely to pick the “right” workout intensity than a human coach or the old algorithm.
• Vision of the Future: The predicted FTP graph for the next month offers massive motivation; you can see how training consistency changes your future fitness number.
3. Handling Non-TrainerRoad Rides (Outdoor / Zwift)
This is the most significant improvement for hybrid training.
• Automatic Ingestion: When you record a ride on a Garmin/Wahoo (outdoors) or Zwift (indoors) and it syncs to TrainerRoad, the AI analyzes the raw file.
• Fatigue & Fitness Impact: It immediately calculates how that effort impacted your fatigue and fitness.
• Immediate Adaptation: If the AI detects high training load from a raw file, it will automatically lower the intensity of upcoming workouts or suggest a rest day.
• No Manual Matching: You no longer need to manually “associate” an outdoor ride with a TR workout to get credit; the system “sees” the work through the raw data.
4. How Progression Levels (PLs) Work Now
Progression Levels have shifted from being the driver of your plan to being a visualization of your fitness.
• “Leveling Up” via Raw Data: The AI analyzes the raw power duration curve of every ride. If a 15-minute effort on an outdoor climb is equivalent to a “Level 5.8 Threshold” interval, your PL updates automatically.
• Resultant, Not Prescriptive: The AI predicts your fitness daily. PLs reflect your current capability based on these predictions rather than just a rigid staircase of workouts.
• Sub-Zone Recognition: The engine recognizes “accidental” work. A long Zone 2 ride with steep hill sprints will update both your Endurance PL and your VO2 Max PL.
• Scoreboard: Progression Levels now act as a holistic scoreboard for all your cycling, not just indoor intervals. You can see your strengths and weaknesses across all riding.
They aren’t taking anything away from the workout player, nor are they forcing you to buy their plans because they are free.
Teach your athletes to pin their workouts /endrant
No Jonathan covered that in a podcast a few weeks back. They did work on it and it did not have any positive impact on the system being able to produce better results.
I’m not certain i understand what you mean here but seems like you mean a platform to just complete and record workouts using a smart trainer. But is providing that and only that even a viable business strategy for 2026?
As far back as March 2022 their website has promoted their offering as providing “Personalized, AI-powered cycling training” (wayback machine web archive: Cycling Planning, Training, & Analytics - TrainerRoad )
I don’t understand how you can be surprised by the growing incorporation of AI.
It’s still possible to create, use and track custom workouts and I guess you, or your clients, can ignore the AI recomendations.
I think this is exactly the opposite of what TrainerRoad is trying to provide.
Instead it’s like having a coach who looks at every workout, and after each workout looks at my performance, analyzing multiple paths forward to find the best outcome for me, according to my goals, and then modifies my training to optimize my results.
They seem to have been deleted. The gist was that they didn’t like the changes, and they made some claims about missing functionality. At least some of the things they wanted (see quotes) have been possible for a long time even before the release of TR AI.
That’s not the vision that @Nate_Pearson has, when Adaptive Training was revealed, he said that in the long term, he wants TR to be better at choosing workouts than coaches and take over the “busy work” of coaching.
We can argue whether it is there yet with TR AI, but coaches like you should IMHO embrace new tech rather than insist on the old ways. It’s like being an accountant and spreadsheets come along. You can look at limitations of early DOS-based spreadsheet software and come to the conclusion “you are doing a better job with your trusty tabulating calculator”, but that’s not sustainable.
Moreover, IMHO your complaint goes in the wrong direction: it seems your athlete trusts TR’s algorithms more than you and/or uses TR to manage their fatigue. Either that means you are not compatible, because you want your athletes to trust your judgement and base their training solely on your recommendations. Or the two of you are communicating past one another.
Why not?
There is loads of similar tech platforms that give athletes input and advice, e. g. the various training readiness scores from Garmin, Fitbit and the like, HRV data and such. TR’s algorithms are merely another addition.
No, if I were you, I’d welcome this. I work in higher-ed and AI is turning the whole system upside down faster than we can come up with ways to adapt. But I don’t think leaning against the tide is a smart approach, you need to lean into it, but pushing in the right direction.
E. g. you come in with the assumption that you are better at picking workouts than TR AI. Well, are you? Be self-critical. Plus, TR AI has capabilities that you might welcome. For instance, if you want to design a custom block of, say, one supra threshold workout, one traditional VO2max workout and 30-30s (not a reasonable example, just making it for the sake of argument), then you can drop these three dynamic workout types in your TR calendar and it will do the rest.
You, the coach, still came up with the design, but the “spread sheet software” TR AI took care of the details. Then, you can still tweak things manually as needed.
I’d also rethink what value human coaches can and do add, things that TR AI and the like can’t (at the moment or in principle). Setting the right goals, separating the wheat from the chaff, race prep and race strategy, monitoring the athlete’s motivation and helping them marry a busy life schedule with training comes to mind.
Except it is the exact opposite of that. When TR started, it was that, you paid for canned training plans that you had to adapt yourself by e. g. choosing alternates manually. That’s where the +/- versions of workouts come from.
Good coaches based their decision on experience having coached many athletes in the past. That’s no match for a computer and a big data set with people who ask the right questions, just how chess computers have long surpassed the best human players.
Since I am not aware of any statistics here, we can argue how TR AI compares to coaches, but if I were a coach (again), I’d experiment with it and see how well it fares against myself.
Having used TR AI for several months now, I can say it is a marked improvement and I haven’t noticed any workout that was clearly wrongly selected. Even when I failed or not passed workouts, there were good reasons outside of TR AI’s control (one time it was oncoming illness, the other time it was that our kids had had a bad morning and were interfering with my training). For reference, I have 8ish years of experience with structured training and briefly coached an athlete.
Get your athletes to turn off these
And disable this
Then get them to manually input whatever FTP number you choose to give them and pin any future workouts you allocate to them, and you’re good to go.
You will, of course, be left with them having to deal with a different coloured background for the 28 days of the simulated calendar.
HTH
Great post
Not happy… been on a structured plan for last 6 months doing fine. Stable FTP 297.
New AI FTP downgraded my FTP to 275…! And says I need do 18 workouts over the next month to get back to 295. Complete nonsense.
I subscribed for a month after a 8 year break from TrainerRoad. Been using Xert ever since. It’s great but struggles with running/weight lifting which I do some of now.
Does the new AI/ML model take runs into account? I can see both weight training and runs in my calendar but I am unsure if this has some impact on fatigue and how the current program I have defined adapts to it. Tried to find information about this but couldn’t find anything.
Does anyone know?
Agree. He was more excellent to the coach than I was for sure.
Some coaching to the coach to think about what their value proposition is.
Post what you’ve done recently and what you have planned. And if you want your TSS history.
The model takes into account their impact on fatigue and adapts accordingly. It does not adapt your running or weight lifting.
In another thread Nate said they want to do more with running in the future
This is a fantastic post, but I also think their whole complaint would be eliminated by simply having their athletes pin their coach assigned workouts and manually setting FTP, wouldn’t it? It would be a terrible use of the athlete’s money, but wouldn’t it work?