Training program design: man vs. machine

What they found

The first key finding was that coaches were generally more positive about the AI-generated programs than athletes who had to actually perform them! Sixty-five percent of the coaches commented that the AI-generated plans were ‘usable’, albeit they needed minor modifications. The athletes on the other hand were far less positive. Only 27.8% found the AI-generated programs worked well for them, while nearly half of the athletes asked their coaches to make major changes to these programs. Meanwhile, 25% rejected them entirely and refused to perform them!

When it came to the suitability of these programs for the distance vs. the sprint swimmers, there was a large discrepancy. The distance swimmers found that while the AI-generated programs worked well for sessions in the low-intensity (A1) endurance/recovery) zone, they struggled significantly with moderate and high-intensity zones (A2, B1, B2), often prescribing too many sets yet insufficient recovery – ie seemingly failing to appreciate the physical and psychological toll of intensive aerobic and threshold training. Conversely, the sprint swimmers commented that while AI suggested an excessive frequency of anaerobic threshold work in the distance swimmers, it prescribed insufficiently frequent sessions in the maximal intensity zones (C1, C2, C3), which are critical for honing sprint-swim performance.

When taking into account the final scoring, it was found that although there was no significant difference in the overall quality of sprinting vs. distance programs, AI produced lower quality scores when trying to prescribe sessions targeting the threshold zones (A2 and B zones), which suggests that it still lacks a nuanced understanding of how to periodize sessions inducing moderate-to-high amounts of cardiovascular demand. The last finding was that older coaches and the male swimmers tended to rate the AI programs more poorly than did the younger coaches and female swimmers.

Implications for coaches and athletes

To answer our initial question about the progress of AI for putting together training programs, it seems that although improvements are being made, there is still plenty of room for progress. This particularly seems to be the case for older and more experienced coaches and athletes, who already have a deeper understanding of what is required to design a successful training program. An AI-designed program also faces another major drawback; while it may be excellent at gathering and synthesizing data, it lacks the coach’s ability to adjust a plan based on an athlete’s immediate psychological or physiological feedback⁽⁴⁾. An experienced coach can read the early signals indicating that an athlete may be struggling with a particular training regimen, and adjust it accordingly. Likewise, experienced athletes can become very good at self monitoring, and know when and how to adjust a program if needed. But when AI is your coach, there’s nobody there to check progress!

In conclusion, this new research suggests that while AI continues to improve, and an AI program may serve as a good starting point - assisting coaches and athletes to get underway by providing a basic template - it will likely need adapting to match the individual physiological and psychological responses of athletes. This is especially true when higher training intensities are required, and where the balance between enough volume to stimulate adaptation and enough recovery times becomes critical for performance. For these kinds of programs, some kind of continual monitoring (through a coach or using wearable technology) should be considered an essential addition to a training template. As the authors eloquently alluded to in their summary, while an AI algorithm can write the ‘music’ of a training plan, it is still up to a human coach or the experienced athlete to ‘conduct the orchestra’ and to ensure the athlete reaches the podium safely and effectively!

References

1.       Business Insider Feb 12, 2026 www.businessinsider.com/microsoft-ai-ceo-mustafa-suleyman-white-collar-tasks-automation-prediction-2026-2

2.       Journal of Sports Science and Medicine (2024) 23, 56 – 72

3.       Biol Sport. 2025 Sep 16:43:355-367. doi: 10.5114/biolsport.2026.152352. eCollection 2026 Jan

4.       Nutrients 2025. 17(3), 442