Training Smarter: Bringing the Uphill Athlete Philosophy to AI Coaching

The shift from “Generic” to “Guided”

The thing with AI, as mentioned in a previous post, is that the output is only as good as the input. When I first asked ChatGPT to create my plan, it did exactly what I asked for – structured, logical, progressive. However, while the plan looked solid on paper, my coach didn’t really understand ME yet.

It knew what to do and how to do it. But it didn’t know why.

It’s at this point that I wanted to change how my AI coach looked at my training. Instead of just writing a plan, I want it to coach like a human would – specifically, one that follows a certain philosophy, which was outlined in The Uphill Athlete.

Discovering The Uphill Athlete Approach

I first came across the idea of the Uphill Athlete while listening to a Freetrail podcast. In the particular episode, the coach of the star performer at UTMB 2025, Tom Evans, went into detail about how he trains his athletes. The philosophy, created by Steve House, Scott Johnston, and Kilian Jornet, as it sounds, is built for more mountain endurance – long, slow climbs, efficient aerobic work, and being smart with your levels of intensity. And while the method is definitely going to help mountain runners, the premise stands true for any ultra-distance runner, even one like me who’s attempting to run a comparatively flat 100-miler.

Why do I say this? Because the demands on the body are similar whether you’re planning to run up a mountain or along the Thames Path. They both require you to be, and stay, in control. You need to be able to manage your effort over time on the course, and whether you are a front-of-the-pack gazelle, or mid to back-of-the-pack runner, you need to build an aerobic engine that keeps you moving at the speed to achieve your time goals.

How This Approach Works For a Flat Ultra

Even I thought that using a framework for stronger mountain running for a flat ultra was slightly odd. But hear me out…the approach, when you think about it isn’t really about the terrain – it’s about efficiency of the body.

What does this mean for me though?

Building a strong aerobic base to be able to stay in zone 2 for longer periods without fatigue.
Controlling the intensity of the run so that I don’t burn out too early in the race.
Developing durability by building strength and consistency rather than speed alone.

It’s not particularly sexy. It’s certainly not about smashing intervals every week (although I will add in intervals as part of the training programme on the lead up to the event). It’s about the discipline to hold back when you want to push yourself harder than you should do. It’s about training at the right intensity, not the highest one.

In an event like the Thames Path 100, the real battles start in the later hours of the race, when fatigue sets in. The discipline to control intensity, and output, matters more than any hill session!

“Teaching” ChatGPT

While ChatGPT can access a plethora of information, it was essential to ensure that it understood what I was truly after. So, rather than just telling it to “follow the Uphill Athlete method,” I fed it the philosophy.

I tried my best to explain what the method stands for: building a strong aerobic base through low-intensity, high-volume training, using the right metrics (Pa:HR (pace-to-heart rate) drift) to monitor efficiency. Focusing on longer-term development, balancing strength, endurance, and recovery to try and improve performance sustainably.

I was also fortunate enough to be able to feed in some recent test results from a VO2 Max & Lactate test I had conducted at Southampton Solent University, which I will go into more detail in a future post. This provided an understanding of how training intensity aligns with my zones:

Zone 1: Recovery – easy pace below 115bpm
Zone 2: Aerobic base – steady running around 120-135 bpm
Zone 3: Tempo/Steady climbing pace – 135-145bpm
Zone 4/5: Threshold and VO2 max work – short, controlled sessions above 145bpm

Once additional information was given to my “coach,” I then asked it to build my weekly structure around that philosophy. Suddenly, the output it gave changed. The suggested workouts looked like:

“65-minute medium-long run. Keep HR <135bpm”
“45-minute easy shuffle. Keep HR <130bpm; focus on relaxed cadence & form”
“50-minute Turbo Z2 spin. HR 120-130bpm”

Instead of just prescribing workouts, it began to coach with a better purpose.

Why this matters

When AI creates a generic plan, it can look fine on paper; however, it will miss the deeper purpose – the intent. Giving a philosophy, such as the Uphill Athlete, brings that intent back.

By teaching my AI Coach how to think like an Uphill Thlete Coach, I’ve said to it, “My goal is not just to survive 100 miles. It’s to arrive at the start line as strong as I can be, run efficiently, and hungry to run the best race I can.”

Throughout the process so far – refining prompts, adding philosophy, providing data – I am learning that AI is no different from having a traditional human coach. The more you tell your real coach, the more specific, the better they can help you.

What’s Next?

So now I have fed my “coach” a philosophy, the next steps, as mentioned above, were to make it even smarter – by feeding in my VO2 max and lactate threshold test results. This will help AI understand my current limits and how to structure better the training intensity around real-life physiology, rather than estimating where I am.

Because if I am asking AI to “coach” me, then I may as well give it as much data as possible to do it properly.

Stay tuned – in the next post, I will share my full test results, what the test involved, how I felt during and after the test, what they mean, and how I am going to use them to shape this block of training.