I’m sold; keep up the good work! It’s kinda janky and slow to respond currently (I am on iOS 27 developer preview) but I expect bugs to shake out as you keep developing. I love the direction you’re going!
Look, I know that AI is relatively non-deterministic, and that makes it a bit...problematic. The subreddit has plenty of documented poor experiences, and I do not take issue with any of them. I just came to report my experience *compared to* the Gemini-based AI Coach in the new Google Health app. (I purchased a FitBit Air.)
I have had both agents create Workouts for me using the exact same prompts for both. And every time, Bevel AI has executed flawlessly--nailed the exercise names, the number of sets, which ones should be labeled warmup sets, the rep numbers (even if they differed between sets) and the weight. Every time. And every time, I have had to go multiple rounds with the Google Health Coach to get the workout right. Sometimes, I've had to give up completely because it just wouldn't save the right exercise, or the right number of sets, or the right weight progression.
Is it because my prompts just happen to be written in a way better understood by Bevel? Maybe. But Google is a behemoth in the AI space. I would expect them to get it right. Indeed, using Gemini alone (outside of Google Health) seems to yield better results, so maybe Bevel workout building functionality is integrated "better" with the AI agent than it currently is in Google Health. Whatever the case, it's been a much better experience so far.