Runner uses AI for training preparation

A runner who had been training four times a week for several years decided to use an AI platform called Gemini to help break their 10 km personal record. Their previous time was 46 minutes and 52 seconds, and they aimed to finish under 45 minutes, ideally 43 minutes, at the 10 km Odyssea in Paris on October 4, 2026.

Training plan adjustments needed

The runner began training in the spring of the previous year, giving them 28 weeks to prepare. They first took a half-Cooper test to determine their VO2 max, a measure of aerobic capacity, and set clear guidelines for the AI to follow. This included asking about the runner’s age, weight, heart rate data, injury history, current training volume, race date, and personal records.

Early challenges with the AI

Initially, the AI proposed training sessions that were too intense for the runner’s current fitness level. One session, for example, required a significant increase in pace and intensity within a short time. The runner had to revise the instructions and add specific rules to ensure the plan was suitable for their abilities. The AI also first suggested a 12-week plan instead of the intended 28 weeks, requiring additional adjustments.

Manual effort and tracking

Once the instructions were corrected, the training progressed well. However, the runner had to manually input sessions into their Garmin Connect app, a process that was more time-consuming than using dedicated training apps. They also converted .fit files into CSV format for analysis, which added extra steps. To keep track of everything, they used a Google Sheets spreadsheet and started new discussions with the AI every four weeks to maintain continuity.

AI adapts to real-life challenges

The AI showed flexibility by adjusting the training plan to accommodate real-life situations, such as a honeymoon trip with elevation gain or days when the runner couldn’t train. Instead of just suggesting to delay sessions, Gemini modified the plan to fit the circumstances. In the final weeks before the race, the AI reduced both volume and intensity to ensure the runner was well-rested and ready for the event.

Race day success and reflection

On race day, the runner started at a target pace of 4 minutes and 17 seconds per kilometer and maintained a consistent effort, following other runners without checking their watch. They finished with a time of 43 minutes and 9 seconds, slightly over their goal of under 43 minutes but still breaking their previous record by 3 minutes and 43 seconds. Despite four short walking breaks, the runner felt confident in their performance.

Future plans and AI considerations

Looking back, the runner expressed some uncertainty about using Gemini, noting that other AI platforms like Claude from Anthropic have since made significant improvements, especially with integration tools for apps like Strava and Coros Training Hub. The runner also mentioned switching between different AIs, including Gemini, Claude, and ChatGPT, to ensure varied training sessions and verify the consistency of each plan. They emphasized the importance of safeguards when using AI, such as ensuring the AI doesn’t push too hard and verifying the appropriateness of each session. For future races, especially a marathon, the runner plans to use a traditional training app to avoid potential programming errors.