Late to cycling: an amateur MTB racer's path to Superfly
Amateur MTB racer, father of three, founder. From random intervals to structured blocks to building Superfly to scratch my own itch.
I’m Vidas. Amateur MTB racer who is also a father of three, a business owner, and one of the two people whose name appears at the top of posts in Field notes. Goal of this post is to give some insight on who’s behind the scenes of Superfly app.
A late discovery
I came to cycling late. Thirty-five years old, after most of an adult life spent doing other things — building a business, raising kids, sitting in too many chairs or actually sitting too much on the chair. I started after the pandemics and enjoyed every bit of riding nature trails. But eventually I got this idea to test myself in racing as I am quite a competitive person.
The first race
The first XC type race was brutal, from having ambitions to finish somewhere around the middle of my age group to grim reality of hoping to finish at all. I bonked, I hiked my bike, I thought I will throw up on every steeper climb. But I survived and it gave me a huge kick in the ass - I must get faster. Silly outcome, but what should I have expected without any training?
Random intervals
To get faster one year was spent just riding and trying to hit those trails as hard as possible. I did progress, felt stronger, got faster on my personal records, but that was not cutting it. I was in the 75% of finishers instead of my initial 50% target. Then next year I read about interval training, a huge kudos for Tom at High North Performance for the amazing content. So I started doing some 4x4minute all out efforts. Sometimes once, sometimes twice a week. Without a power meter.
Of course aerobic capacity went up — anything goes up when you start from a low base — and the gains felt large because they were the easy ones. The pattern’s flaw shows up later: every session is medium-hard. The mid-zone gets ridden constantly, easy days are not easy enough to recover from, and hard sessions are not hard enough to drive adaptation. Stöggl and Sperlich’s polarized training research is one of several studies showing this is the gravity well almost every self-coached athlete falls into.
But such training was very hard to keep up as I felt exhausted most of the time. Even though I was getting closer to my 50% group finishers target, I figured that I do a lot but gain small.
Structured blocks
I finally got a power meter. Then the next layer was a plan-template subscription. Eight-week base, build, taper, race. The structure helped — somebody else picked the sessions, somebody else placed the recovery weeks, the medium-hard temptation receded. The frustration was different. The plan didn’t know that one of the kids had been up at 4 a.m. coughing, or that a client meeting had moved the only available indoor session, or that the last testing already invalidated this week’s intensities. It marched forward. I marched along behind it, increasingly out of register with what I should actually be doing on any given day. It felt frustrating again.
AI-assisted plans
The next step was the obvious one for someone who builds software: use AI to write the plan. I wired up a working setup with a large language model on top of my Strava history and let it produce week-by-week prescriptions. It was promising. The agent could read a week of activities and adapt next week’s load. The conversation was fluent. The plan was passable.
It also kept hallucinating physiology. It would silently rearrange a polarized week into a threshold-heavy one. It would ignore a testing update I had just submitted. It would recommend a recovery week one cycle early, then a build block immediately after, when the metrics said the opposite. The shape of the answer was right; the substance underneath was unreliable. I do not want my training derived from a system that needs a coach watching it to keep it safe. Generative coaching from a generic model is a deceptively good demo and a poor long-term plan.
Scratching my own itch
What I wanted by then was specific. A real adaptive and well tested training. I was lucky to meet Andrejus who is in my eyes a semi-pro athlete with deep understanding for training. Andrejus poposed a method grounded in lactate thresholds and the Norwegian endurance model. Method which is executed by software that adapts when life moves things around. We also wanted to have a software that is accompanied by something that can answer questions 5:30 AM in the morning or late in the weekend - an AI companion. But the methodology and training part remain human crafted.
So my co-founder Andrejus and I started building it. That is Superfly. The methodology comes from a coach who races, written by hand. The software handles scheduling, Strava matching, multi-season CTL/ATL/TSB tracking, and the small adaptations that keep a real human’s training plan honest. The AI companion answers questions inside a plan — it never generates one. The full case for that division is in the companion post on why Superfly exists. This one is just the personal half.
My key take aways
- Structure and consistency is of extreme importance where a lot of amateur athletes fail.
- Easy rides are the hardest to do. I had to learn how to ride easy, as I was used to attacking trails all out every time.
- Tracking own progress over years helps to see the big picture. I will blog about it separately.
Why I write here
Most of what gets shipped in software ships without explanation. Training is the opposite — a programme is half method, half judgement, and the judgement degrades when it isn’t written down. Field notes is where the judgement parts get written down: why a recovery week sits where it does, why an aerobic-threshold session is the duration it is, why we avoid the mid-zone almost religiously.
I am also the first reader for everything published here, in the literal sense that I am the athlete the templates were originally designed for. If something on this blog turns out to be wrong, that is on me first. The arc, in short: late to cycling, three kids, a business that does not stop because the legs are sore, three years of progression that taught me what structured training is actually for, and a tool built because none of the existing options fit the shape of an amateur racer’s life.
FAQ
Did starting cycling at 35 hurt the long-term ceiling?
Less than people assume. The aerobic adaptations endurance training drives — mitochondrial density, capillary network, stroke volume, lactate clearance — respond to training load at any age past adolescence. What a late starter loses is the years of base volume that early-starters bring into adulthood, which means the first three to five seasons compound faster than later ones. The ceiling is real, but it sits dramatically higher than the one most amateurs reach before they get bored or quit.
Why didn’t AI-generated training plans work for you?
The shape was right; the physiology underneath was unreliable. The model would silently rearrange a polarized week into a threshold-heavy one, ignore a testing update I had just submitted, or place a recovery week to optimise a metric rather than to recover the athlete. Generative coaching from a generic model produces a confident-sounding plan that needs a coach watching it to keep it safe — which defeats the point of the automation in the first place.
How do you train as an amateur racer with three kids and a business?
You ride at the edges of the day. The 10 p.m. trainer session, the 6:30 a.m. ride before the school run, the compressed Saturday because the oldest is in tournaments. The methodology has to assume the calendar will move and the load has to compound across years rather than weeks. That is the design constraint Superfly was built around.
What does an amateur MTB racer’s winter training week look like?
Roughly: Z2 sessions and a bit of body work. But it’s booring indoors so I try to extend my season to the very limits.
Is Superfly only for mountain bikers?
No. The platform supports road, gravel, mountain bike, cyclocross, and track cycling. The methodology — lactate-threshold structure, Norwegian polarization, multi-season periodization — is the same across disciplines. The workout content adjusts to the demands of the event you are training for.