If you spend any time around fitness apps, you'll notice they all do approximately the same thing. They give you a place to record what you did, when you did it, and how heavy or how long. The interface changes — Strong is clean, Hevy is generous, MyFitnessPal is a search box plus a database — but the model is the same. They're filing cabinets. Useful filing cabinets. But filing cabinets.

We've built something different. Not a better filing cabinet. Not a filing cabinet with AI bolted on top. We built a coach: something that drafts your week, asks why you skipped Tuesday, swaps a squat for a hotel-gym alternative, and remembers — not just the data, but the goal. And then we built it to talk.

This essay is a long version of why.

The category that already happened

Fitness software has gone through three phases, roughly.

Phase one was tracking. The 2010s were dominated by apps that turned your phone into a notebook. Strong, FitNotes, Hevy. Each set, each rep, each weight, recorded by hand. Calorie tracking apps did the same for food. The promise was that the act of writing it down would teach you something — and that the data would compound into insight.

For a small population, it does. For most people, it doesn't, because logging is friction and friction kills habit. The macro spreadsheet quietly stops getting filled in. The workout app sits unused for a month. The data is partial, then misleading, then abandoned.

Phase two was planning. Apps that didn't just record what you did but told you what to do next. Future, Caliber, the better personal-training apps. Sometimes a coach was a real human on the other end; sometimes it was a programmed routine. The promise was specificity — a plan made for you, not pulled off a shelf.

This works, when you can afford it. Personal training is the second most effective fitness intervention after the surgery you don't want. But it costs $200–$500 a month, and the entire pitch is hard to scale: a coach has hours, and hours have prices.

Phase three is conversation. Not tracking. Not pre-planned routines. Not even a remote human checking in once a week. A model that drafts your week, listens to what's actually happening, and adjusts in language.

This is what large language models, deployed carefully, finally make possible. It's the bet behind Kin.

What "tracking with AI" gets wrong

A lot of apps in 2025 and 2026 will tell you they have AI now. Most of them mean: we added a chatbot to a tracker.

This isn't nothing. It's not Kin, either. The shape of the product matters. If the chatbot is a side door — a thing you open from a settings menu to ask a question — the app is still a tracker, with a help feature. The data still flows the old direction: you input, the app stores. The relationship is the same.

"AI tracker" is a category mistake. A tracker stores. A coach plans, adapts, remembers, explains. Adding a chatbot to a tracker doesn't change what the product is.

What we wanted instead: an app where the conversation is the product. Where you don't open a tracker and then summon an assistant. You open a coach. The plan you see, the workout you log, the meal you ate — they're all part of one ongoing exchange. The data flows both directions.

This is harder to build, because everything has to be designed around the conversation rather than around forms. But it's also the only way you get a thing that genuinely behaves like a coach instead of an app with a chat button.

Why Claude, specifically

We built Kin on Anthropic's Claude. People sometimes ask us why — most LLMs can answer fitness questions in a vacuum.

The honest answer is that there are three properties we needed, and Claude does them well.

Long context. A real coaching conversation doesn't reset every message. If you mentioned six weeks ago that you were rehabbing a shoulder, a coach should still know that. Claude's long context window means we can keep your goals, recent sessions, and adjustments in mind without summarizing them into a flatter, blurrier version.

Following instructions. An off-the-shelf chatbot will tell you what it knows about training. A coach has a specific way of working — when to push, when to back off, when to ignore what you said and ask the better question instead. Claude is unusually good at following the kind of layered instructions that turn a model into something that behaves consistently in a specific role.

Restraint. You don't want a coach that hallucinates a sports-medicine paper to back up a bad take. You want one that says "I don't know — let me ask you a clarifying question." Claude is, by design, more willing to admit uncertainty than most models. That property is essential when health is involved.

We also liked, and we'll just say it, that Anthropic doesn't train on customer API data. Your conversations stay yours.

What Kin is not, and won't pretend to be

A few things we want to be honest about, because the AI fitness category is going to fill up with apps making bigger claims.

Kin isn't a doctor. It won't tell you whether your knee pain is a meniscus issue. It will tell you to see a physio.

Kin isn't a substitute for a great human coach if you have access to one. A 1:1 relationship with a strength coach who watches you squat for an hour a week is still better. We're trying to give the millions of people who don't have that something better than a tracker.

Kin doesn't pretend that an LLM has read every paper. It has knowledge boundaries, and we surface them. When it doesn't know, it says so.

Kin won't try to be a comprehensive medical record, a sleep coach, a meditation app, and a wearable hub all at once. It does training and nutrition coaching. The product is allowed to have an edge.

The shape of the bet

Here's what we believe, in plain English.

The 2020s for fitness will look the way the 2010s looked for finance. The category got dominated, then re-defined, by software that wasn't trying to digitize an existing relationship — it was trying to deliver, at software prices and software scale, a service that previously belonged to a small group with money and access.

Personal training is that service for fitness. It's effective, it's expensive, and most people who would benefit from it never get it. The wedge isn't going to be a slightly cheaper personal trainer. It's going to be a different kind of product entirely — one that delivers most of what coaching does, conversationally, on demand, for the cost of a streaming subscription.

That product is what we're building. It's why the homepage doesn't say "AI workout tracker." It's why the hero film opens with the word listens.

The closer you look at the product, the more you'll notice what isn't there: the dropdowns, the eight-tap entry, the tabs labeled "stats" and "history" and "exercises." We took those out on purpose. The conversation is the surface. Everything else is a side effect.

That's the bet. We think you'll like the way it feels.