Nathan Limm is putting AI to the test. Over 10 weeks, Microsoft Copilot is training him to take 15 minutes off his half-marathon time, using information on his physiology and daily training logs. Success could point to potential for a generation of top athletes to be coached by AI.
Withthe meteoric rise of artificial intelligence, industries worldwide are still getting to grips with how best to utilise it. Part of what makes artificial intelligence (AI) so revolutionary is its accessibility. Anyone with an internet connection can tap into it. AI is being used in both professional and amateur contexts to help humans achieve goals. But what about in the sporting world?
It’s not hard to imagine a young athlete – hell-bent on achieving greatness in their sport, but without access to coaching – asking for tips, tricks and training programmes from a digital companion. But how effective will the advice be? Could we have a generation of All Blacks and Olympians trained by AI?
AI has been used in professional sports environments to varying degrees of success. Ex-Spain football manager Roberto Moreno was sacked as coach of Russian club FC Sochi in September after using ChatGPT to create a travel and training plan for the team. Moreno failed to adapt the schedule when the AI suggested the team stay awake for 28 hours during an away stint.
In contrast, women’s football coach Laura Harvey took the Seattle Reign from 13th in 2024 to fifth in 2025 in the National Women’s Soccer League after she started changing the team’s formation based on ChatGPT’s suggestions for each opponent.
Nathan Limm hits the training track on a half-marathon preparation plan devised by AI. Photo / Annaleise Shortland
Over 10 weeks, I’m putting AI to the test. I’ve tasked Microsoft Copilot with training me for a half-marathon on Auckland’s waterfront on April 12. The ambitious goal is to reduce my time from 1hr 55m to 1hr 40m.
I’ve run marathons before, but I’m not a professional athlete. I work full time, I have a social life, and I have a minor back issue, all of which the AI will factor into the training.
Turning the AI into a running coach wasn’t as simple as typing in a one-sentence prompt. In fact, AI played a role in helping turn it into a coach. I tasked it with creating an optimal prompt for the project, and it asked for information on my physiology, fitness, lifestyle, goals, available facilities and equipment, and how I’d like to be held accountable.
Nathan Limm keeps an eye on his time on a half-marathon preparation plan devised by AI. Photo / Annaleise Shortland
It then gave us a prompt to copy-paste into a fresh Copilot chat. From there, I could interact with it as I would a real running coach.
Before starting the programme, I ran a half-marathon to get an idea of a starting point, and my time was 1h 55m. Based on this time, 1h 40m is an ambitious goal, one that will test both my resilience and the AI’s nous.
Only the bare basics in equipment will be utilised. I use a free app, Strava, to track my runs. I have no fancy watch and no heart-rate monitor. This means pacing is used to measure the intensity of each workout.
I’ve asked for a 10-week programme, and AI is giving me the schedule week by week. The programme is adjusted based on my daily logs, in which I tell Copilot about my sleep, fatigue, soreness and back symptoms.
Week one: Letting down my coach
Guilt. Legitimate guilt. It’s odd that only after a few days, I’d formed a relationship with my AI coach strong enough to feel like I’d actually let it down. I’m aware it sounds ludicrous, but the detail, attentiveness and encouragement I receive from the bot have led me to genuinely trust it.
I’m using Microsoft’s AI, Copilot, simply because it’s the one we use at work. My plan to get Copilot to train me over 10 weeks to run a half-marathon in less than 1hr 40m in April has started slowly. I returned from overseas travel with a cold and low energy levels, and the AI has responded accordingly. The gentle start was peeled back further to prioritise my recovery. Runs were replaced with mobility sessions to manage my back niggle, and any time my fatigue crept up to a 4 out of 5, a full day of rest was mandatory.
Although I felt worried that the entire plan could be compromised by this cold, my coach reassured me that rest would protect the first block of the programme, and a sub-1h 40m half-marathon is still very much on the cards.
By Saturday, I had recovered enough to go for a light, 25-minute jog to shake off the cobwebs. The run – probably more accurately described as a meander – was hard. Slowing down to a walk at times, my energy hadn’t fully returned, and I felt beat after just 23 minutes and 3.38km. Fatigue lifted from a 2.5/5 to 3.5/5, and my back issue flared a little. My coach, however, was fairly encouraged. The initial 8km run we had discussed for Sunday was downgraded to 5km, but Copilot said I’d handled the conditions perfectly.
Nathan Limm working hard on the Lovelock Track, in Auckland. Photo / Annaleise Shortland
There was, however, a warning. Copilot said my rising fatigue was my body saying: “Thanks for the stimulus … don’t add more today.” Unfortunately, I had other plans. Months ago, I’d bought tickets to Worship, a music festival in the Auckland Domain for that night. The tickets were $180. Was I going to let an AI tell me not to go? Like hell.
I was like a high school rugby player going to a party the night before a game. Without consulting my coach about the potential ramifications, I headed to the festival and spent hours on my feet, dancing in the mosh. She’ll be right, right?
I woke up the next morning feeling grim. I gingerly pulled out my phone and confessed to my coach that I had spent the night out, and my fatigue had spiked to a 4/5.
Already knowing I had made a mistake, the “good on you for being honest” I received from Copilot felt like a “I’m not angry, just disappointed” response from my parents.
Call me a Gen Z snowflake, but when your entire interaction with your coach is in text form, you become quite sensitive to the number of exclamation points, indicating tone.
The 5km run was scrapped, and I spent the day in bed, feeling guilty but determined to recover to log a more productive week in the coming days.
Copilot could have been a lot harsher, as it did reassure me I was doing the right thing by resting and protecting week two. However, anyone taking sport seriously is already imposing sizable self expectation, and anything other than success will feel disappointing.
It’s definitely ridiculous, but there’s a small part of me that wants to make my non-existent coach proud.
Nathan Limm has been a journalist with Newstalk ZB and the NZ Herald since 2020. He covered the Netball World Cup in Cape Town in 2023, hosts The Big League Podcast and commentates rugby and netball for Gold Sport.