amp AI Coach

The Problem

When I joined the company, amp's home screen worked like Netflix, a wide catalogue, browse by mood in the app's home screen. Research and beta feedback kept pointing to the same gap: users didn't want more content. They wanted to feel guided. The business had a ceiling, too. Trainer contracts and content shoots were expensive, slow, and impossible to personalize at scale. The question became: how do you build something that feels like a real coaching experience without the constraints of one?

The Problem

When I joined the company, amp's home screen worked like Netflix, a wide catalogue, browse by mood in the app's home screen. Research and beta feedback kept pointing to the same gap: users didn't want more content. They wanted to feel guided. The business had a ceiling, too. Trainer contracts and content shoots were expensive, slow, and impossible to personalize at scale. The question became: how do you build something that feels like a real coaching experience without the constraints of one?

The Problem

When I joined the company, amp's home screen worked like Netflix, a wide catalogue, browse by mood in the app's home screen. Research and beta feedback kept pointing to the same gap: users didn't want more content. They wanted to feel guided. The business had a ceiling, too. Trainer contracts and content shoots were expensive, slow, and impossible to personalize at scale. The question became: how do you build something that feels like a real coaching experience without the constraints of one?

Defining the Trainer

I pushed to bring a physical trainer into the office. I interviewed him, watched him work, and did a full session myself. Four things stood out.

Before the session: how did you sleep? How sore are you? He adapted on the spot without losing sight of the goal.
In-session: feedback. Real-time output monitoring. Too easy - he pushed. Too heavy - he adjusted.
After the workout, he told you how you did. A human signal that the work mattered.
Beyond the session: Recovery, nutrition, next steps. Coaching didn't stop when the workout ended.

These four pillars became the design framework for everything that followed.

Defining the Trainer

I pushed to bring a physical trainer into the office. I interviewed him, watched him work, and did a full session myself. Four things stood out.

Before the session: how did you sleep? How sore are you? He adapted on the spot without losing sight of the goal.
In-session: feedback. Real-time output monitoring. Too easy - he pushed. Too heavy - he adjusted.
After the workout, he told you how you did. A human signal that the work mattered.
Beyond the session: Recovery, nutrition, next steps. Coaching didn't stop when the workout ended.

These four pillars became the design framework for everything that followed.

Defining the Trainer

I pushed to bring a physical trainer into the office. I interviewed him, watched him work, and did a full session myself. Four things stood out.

Before the session: how did you sleep? How sore are you? He adapted on the spot without losing sight of the goal.
In-session: feedback. Real-time output monitoring. Too easy - he pushed. Too heavy - he adjusted.
After the workout, he told you how you did. A human signal that the work mattered.
Beyond the session: Recovery, nutrition, next steps. Coaching didn't stop when the workout ended.

These four pillars became the design framework for everything that followed.

Designing Within AI Constraints

AI-generated video and audio are still experimental and expensive. A daily coach video wasn't feasible. So I treated the constraints as an opportunity. Every Sunday night, the system generates a weekly coach video setting focus and intention for the week ahead. Users ended up coming back to it multiple times. For daily context, audio notes were faster and cheaper to produce, so each workout got a short personal brief from the coach, just like a WhatsApp voice message.
Weekly video for the big picture. Daily audio for the moment. Two assets, two jobs, one continuous coaching presence.

Designing Within AI Constraints

AI-generated video and audio are still experimental and expensive. A daily coach video wasn't feasible. So I treated the constraints as an opportunity. Every Sunday night, the system generates a weekly coach video setting focus and intention for the week ahead. Users ended up coming back to it multiple times. For daily context, audio notes were faster and cheaper to produce, so each workout got a short personal brief from the coach, just like a WhatsApp voice message.
Weekly video for the big picture. Daily audio for the moment. Two assets, two jobs, one continuous coaching presence.

Designing Within AI Constraints

AI-generated video and audio are still experimental and expensive. A daily coach video wasn't feasible. So I treated the constraints as an opportunity. Every Sunday night, the system generates a weekly coach video setting focus and intention for the week ahead. Users ended up coming back to it multiple times. For daily context, audio notes were faster and cheaper to produce, so each workout got a short personal brief from the coach, just like a WhatsApp voice message.
Weekly video for the big picture. Daily audio for the moment. Two assets, two jobs, one continuous coaching presence.

Where the Coach Meets the User

The home screen shifted from a content catalogue to a focused weekly view. Your week, your workouts, your coach. Everything else is removed from the primary surface. The coach got a dedicated area front and center: weekly video, fitness scores, insights, and direct interaction. One place where the relationship lived. The message changed from 'here's what you can do' to 'here's what you should do, and why.' The next step on the roadmap is a periodical training view: a multi-month perspective showing where a user sits in their program cycle. Every design decision was made with that next step already in mind.

Where the Coach Meets the User

The home screen shifted from a content catalogue to a focused weekly view. Your week, your workouts, your coach. Everything else is removed from the primary surface. The coach got a dedicated area front and center: weekly video, fitness scores, insights, and direct interaction. One place where the relationship lived. The message changed from 'here's what you can do' to 'here's what you should do, and why.' The next step on the roadmap is a periodical training view: a multi-month perspective showing where a user sits in their program cycle. Every design decision was made with that next step already in mind.

Where the Coach Meets the User

The home screen shifted from a content catalogue to a focused weekly view. Your week, your workouts, your coach. Everything else is removed from the primary surface. The coach got a dedicated area front and center: weekly video, fitness scores, insights, and direct interaction. One place where the relationship lived. The message changed from 'here's what you can do' to 'here's what you should do, and why.' The next step on the roadmap is a periodical training view: a multi-month perspective showing where a user sits in their program cycle. Every design decision was made with that next step already in mind.

Outcomes & Impact

AI-generated workout adoption increased 30%. Library tab usage dropped 11%. Users stopped browsing and started following a plan. More telling: the nature of user feedback changed. Before, people asked for clarity and guidance. After launch, customer feedback faded in that area. The new requests are about customization and more control over their plan. That shift means the foundational problem is solved. We moved users from confusion to confidence, and now we're building the next layer on top of that.

AI-generated workouts

+30%

Library tab usage

-11%

Outcomes & Impact

AI-generated workout adoption increased 30%. Library tab usage dropped 11%. Users stopped browsing and started following a plan. More telling: the nature of user feedback changed. Before, people asked for clarity and guidance. After launch, customer feedback faded in that area. The new requests are about customization and more control over their plan. That shift means the foundational problem is solved. We moved users from confusion to confidence, and now we're building the next layer on top of that.

AI-generated workouts

+30%

Library tab usage

-11%

Outcomes & Impact

AI-generated workout adoption increased 30%. Library tab usage dropped 11%. Users stopped browsing and started following a plan. More telling: the nature of user feedback changed. Before, people asked for clarity and guidance. After launch, customer feedback faded in that area. The new requests are about customization and more control over their plan. That shift means the foundational problem is solved. We moved users from confusion to confidence, and now we're building the next layer on top of that.

AI-generated workouts

+30%

Library tab usage

-11%