Whoop CTO on AI in Fitness, What’s Next for Coach

ATN spoke with Jaime Waydo about Whoop’s new advanced generative AI tool and its plans around artificial intelligence

Whoop recently broke new ground in the wearables industry with the launch of Whoop Coach, an advanced generative artificial intelligence (AI) feature that uses GPT-4 technology to deliver highly personalized health and fitness coaching on demand.

Created in partnership with OpenAI, the lab behind ChatGPT, the AI tool leverages Whoop proprietary algorithms, a specially designed machine learning model and a member’s unique biometric data to help users better understand their Whoop scores.

Athletech News spoke with Jaime Waydo, the chief technology officer at Whoop, to better understand why the wearable health and fitness company launched Whoop Coach and learn more about how it plans to use AI moving forward.

This conversation has been lightly edited for clarity and length

Athletech News: What prompted the launch of Whoop Coach? 

Jaime Waydo: I joined Whoop a little less than a year ago, and from the very beginning, we were talking about how to use more and more AI in the product. When OpenAI announced Chat GPT-3 at the end of November 2022, you could have a conversation with it: you could ask it questions, and it would respond. You could refine those answers. That had never really happened before. I sent a message on December 1st to the whole company, and said, “If you’re not playing with this, check it out.” At the time, we had just finalized the 2023 roadmap for Whoop. I typed into chat GPT-3, “What should the roadmap for Whoop be for the next year?” It came back with some really good ideas—things that we had already thought about. And that’s when I knew: there’s something really powerful here if we can tap into it.

So then the conversation became: what happens if you tie Whoop data into a large language model? What’s the conversation that can happen? And is that something that starts to feel like that 24-7 world-class performance coach, which is a lot of Whoop’s vision. In early January, the team started playing with large language models. Then we really started a few different explorations. One of them was figuring out the voice of the brand. A lot of work also went into tying it together with all of your data and making sure that no matter what question you ask, Whoop Coach has access to the right pieces of data to be able to give you not just an answer that’s obvious or that you could find on Google. We spent a lot of time figuring out what the interaction would be like. It was fun because this technology has never existed in the world before.

We launched it in September and the feedback so far has been hugely positive. The team’s off working on the next iteration of Whoop Coach that will be launching in a couple of weeks and we’ll continue to iterate on it and train it on more data and more models, and it’ll just get better and better and better every week.

ATN: What are the most popular searches that people start with on Whoop Coach? 

JW: There are a couple of emerging themes that I see. One is for new members on Whoop. That’s questions like “What is strain?” or “How do I get in the green tomorrow?” Then, there’s what I call the data enthusiasts. These are the people who want to know how this data compares to that data. So, “Do I sleep better in the summer versus the winter?” These are things that if you look in the app, it’s really hard to figure out. But now you can ask Coach and at your fingertips, you’ll get that answer. The third emerging theme that I see is around what I call the “time optimizer,” the person who just wants you to tell them what to do. “What’s the workout I should do today?” or “Give me a plan to run a marathon in two weeks.” 

ATN: Has the Whoop team thought about making the voice of Coach customizable? 

JW: Yes, there’s a set of technology that we’re working on right now where the coach learns the way that you respond best. Maybe as a woman you respond better with a particular voice in certain phases of your cycle or you need a specific voice during recovery days. The team is experimenting with all of that right now and working on that next layer of customization.

ATN: How do you see Coach changing the world of strength training for Whoop? 

JW: We’re excited about connecting Strength Trainer to Whoop Coach. Whoop Coach could recommend specific workouts depending on your strain goals for the day. One of the things we talk about a lot at Whoop is, How do we weave things together? So, when we build Coach, we don’t just build Coach. We talk about, How does Coach work with Strength Trainer? How does Coach work with sleep? How does Coach work with the journal? All needs to work as one. So we work very hard on making sure that as we’re designing Coach, we’re designing how to tie it into the journey.

ATN: How have you navigated potential regulatory challenges when building out Whoop Coach? 

JW: What we care about more than anything is privacy and security. We make sure that no one’s personal data leaves the four walls of Whoop. Even though we’re partnered with a large language model provider, we still make sure that all of your personal information only stays at Whoop and no one’s accessing that data. Open AI was a great partner for us; they don’t retain anything that we send over, they don’t train on any of the data, so that’s been a great foundation for us. 

ATN: What’s next for Whoop Coach? 

JW: One of the things that’s true of all AI technology is that you’re always training on more data so that it’s more capable. There are certain questions that if you asked Whoop Coach, it would probably say, “Our engineers are hard at work adding that data so that Coach has access to it.” For example, if you asked Coach right now, “What’s your battery state of charge?” it can’t tell you that, but soon it will be able to. There’s a lot more performance science that we are integrating, so it will give you deeper scientific insights. It will also weave into other parts of the app, which I think is very important.