This is both a blessing and a curse of AI.
Quality of data matters immensely, both in training, as well as for the individual user. Garbage in = garbage out.
And unfortunately, a gigantic portion of the puzzle that remains largely non-objectively defined, is how you ‘feel overal’', which is obviously super subjective, but super important. I’ve had periods of overtraining where I was crushing every workout, setting shorter race PRs left and right, but knew something was ‘off’ as I didn’t feel good at all, and had that ‘fried’ feeling day in and day out, but was too inexperienced back then to back off from all those tempting PRs. Then comes the massive underperformance on A race day despite all that heroic training. Maybe AI will get better at interpreting subjective descriptions of it and take it into account, but I suspect it’s already going to be quite a challenge for AI to just outperform well-designed stock programs (like 80/20 plans) that are just incrementally a step up from your current level.
ChatGPT already can generate really generic, basic plans that have nothing wrong with them, even for a 24 week ironman build. I’ve seen them, and I’d absolutely have solid results with them, with no editing required. It’s that extra small percent of performance that you want to squeeze out of training that’s the challenge, as it’s quite individual.