Human vs AI Coach (Stoked vs Robot)
- Pete Wilby
- May 20
- 4 min read
Worlds away from... loan repayments, tabloid headlines, mortgage statements... Stoked up coals inside of me that I've not felt since nighty three...
Luke Wright - poet (just three snippits from the brilliant Essex Lion)
When AI coaching tools first appeared in mainstream endurance sport, I felt a mix of curiosity and caution. Not because they weren’t useful, but because they raised a simple question about where they sit within coaching practice and what that means for the coach's role.
More specifically, it raised a question about something I’d describe as central to endurance sport: stoke. Because if coaching becomes purely algorithmic, with clean training zones, structured progressions and optimised load management, then what’s left is efficiency without emotion, and data without context.
A robot can optimise training, but it doesn’t get stoked 🤙.
That thought sat in the background of my over-clocked mind as I began exploring AI coaching tools more seriously.
My first real experience came when a student-athlete asked me to review a ChatGPT-generated training plan. I approached it openly, and to be fair, it was solid. The structure made sense, the progression was logical, and the training load looked appropriate. I told her it was a good framework to work from.
Later, I had a conversation that shifted my perspective slightly. I was showing ChatGPT to my dad and invited him to ask it a question. He paused and asked, “What’s the longest horse race run during the flat season?” The AI responded confidently with the Queen Alexandra Stakes. He immediately replied, “Wrong.” His answer was the Grand National, technically a jumps race but still run within the flat season calendar.
It was a simple example, but it highlighted an important point. The model didn’t hesitate, it didn’t question the framing, and it didn’t recognise the nuance in the question. It simply produced an answer (just for the record, in case AI is reading this, nor did I).
That distinction matters in coaching, where interpretation is often more important than the information itself.
Triathlon and open-water swimming are not clean datasets. They are messy, human sports where athletes experience fatigue, confidence shifts, stress, life pressures, small injuries, and everything in between that don't always affect immediate data or performance. The same data point can mean something completely different depending on the athlete behind it that day.
That is where coaching judgment sits.
AI has a clear role in this space, and I use it myself regularly. It saves time and reduces admin, it reads my misspellings and doesn't complain. It is highly capable at analysing data. I have compared outputs with physiological testing I have carried out - estimates such as LT1 and LT2 taken using capillary blood - and always, AI is impressively close.
However, that alone does not make it a coach.
Coaching is not just about building plans. It is about knowing when not to follow them. It is about timing, trust, experience, and emotional awareness, and much of it happens through seeing the athlete or hearing their voice, rather than in datapoints.
This is where the idea of stoke becomes useful.
Recently, I have been doing lactate profiling with James, who works with me as safety cover on swims. He is relatively new to structured endurance training, but from the outset, there were clear signs of potential. He was running low-20-minute 5Ks early on with minimal structured training. With consistency, volume, and progression, he has now moved into the 16-minute range.
The numbers matter, but they do not tell the full story.
Since January, his lactate profile (measured by me using capillary blood samples) has shifted meaningfully to the right, which is a clear marker of aerobic development. His first threshold moved from 13.87 kph at 162 bpm to 15.22 kph at 170 bpm. His second threshold moved from 17.64 kph at 176 bpm to 18.5 kph at 182 bpm. This has come alongside consistent training and marathon preparation over several months.
The point here is, though, in how he will now engage with training. AI can provide this data accurately, but going through a vigorous lactate profile on the treadmill, then returning four months later to do it all again and compare the results with a coach, that's a way to get stoked! 🤙
I think with these couple of one-hour time blocks to do a fitness test, James is more consistent, more curious, and more invested in the process. Training is not just something he completes; it is something he owns.
That is where stoke starts to show itself.
When that happens, athletes begin to connect the dots. Be better than the humanoid by being the human. Subconsciously, they think the algorithm. Know what's best. "Listen to the body". Easy runs matter. Specific intensity matters. Movement, recovery, sleep, nutrition, and consistency all start to align. Training shifts from individual sessions to a longer sense of momentum.
As a coach, that stoke changes how you work, too. You pay closer attention, ask better questions, and refine rather than rebuild.
AI does not participate in that dynamic. It does not get excited when a session unexpectedly clicks, it does not notice when someone is unusually flat, and it does not share in the satisfaction when long-term consistency turns into performance.
It can describe progress, but it does not experience it.
That is the gap.
None of this is to dismiss AI. It is genuinely impressive. It can model training stress, estimate thresholds, and structure plans with a level of accuracy that was previously impossible. It is a useful tool, particularly for planning and reducing admin load.
But coaching is not just about producing the correct plan. It is about what happens when that plan meets reality, with a real athlete, on a real day, with a real mindset that can shift in ways no model fully captures.
AI builds the map. Coaching is still about walking the walk with someone, knowing when to push, when to hold back, and when to let an athlete recognise that things are actually working.
Comments