Stride vs. Building Your Own with Claude or ChatGPT

"Couldn't I just build this myself with Claude or ChatGPT?" Yes, you could, and if you enjoy that sort of thing we would encourage you to try. We use AI tools every day ourselves. They make a good engineer faster. They don't tell you what to build, or how to build it so that it keeps working when a thousand things upstream go wrong.

Stride is engineered by people who build software for a living, and it has been in development for well over a year. What we can tell you is where that time goes. Getting a sensible training plan out of a model takes an afternoon. Everything around the model is the real work: the integrations, the data pipeline, the sports science, the native apps and the infrastructure that keeps it all running.

What the first afternoon gets you#

Export a few rides, paste them into a chat, describe your goal and ask for a plan. You will get something convincing: sensible weeks, a build, a taper, some intervals. For plenty of people that is enough, and it costs nothing beyond a subscription you already have.

The trouble starts on day two. The plan doesn't know what you did yesterday, how you slept, or that Thursday's session never happened. So you paste more in. Then you want it to happen on its own, and that is the point where you stop writing prompts and start writing software.

Getting your data in, all of it, every day#

A useful coach needs every ride, run and swim, plus your sleep, HRV and resting heart rate, arriving without you lifting a finger. That means a separate integration for every service, each with its own sign-in, webhooks, rate limits and file formats. Stride maintains them for Garmin, Wahoo, COROS, Polar, Hammerhead, Zwift, Rouvy, Whoop, Oura, Apple Health and Health Connect.

Connecting them is the easy bit. Keeping them working is the actual job. In practice, that has meant:

• The same ride arriving three times. Record on a Garmin, ride in Zwift, sync through Wahoo, and you have three copies with slightly different start times. Stride has to keep the best one and merge the rest, or your training load triples.
• Timezones, everywhere. Devices record local time in different ways. Get it wrong and a morning ride in New Zealand lands in the future, or last night's sleep is filed under the wrong day for everyone east of London.
• Files that aren't what they claim to be. One indoor platform writes dates twenty years in the future. Some gym equipment exports files that aren't valid FIT at all. E-bike rides turn up labelled as something other than cycling.
• Connections that die quietly. Access to a provider expires and nothing tells you. Your data just stops, and you notice a fortnight later when the plan has gone stale.

Sources also disappear. On 1 September 2026 Strava changed its terms to stop apps passing athlete data to an AI, which closed off the most obvious shortcut for a home-built coach. See Stride and Strava.

Numbers you can trust#

Language models are fluent with numbers they haven't actually calculated. Paste in a power curve and ask for your FTP, and you will get a confident figure. Whether it is right is another matter.

Stride computes its numbers from the raw file, second by second, rather than asking a model to estimate them: normalised power, W′ balance, aerobic decoupling, interval and sprint detection, power curves that don't stitch efforts together across a pause, and weather along the route you actually rode. The AI then reasons over those numbers instead of inventing them.

We also test them against real athletes. When we checked the common rule of thumb for estimating FTP against riders' actual tested efforts, it came out around 27 watts low on average, enough to make every threshold session too easy. That kind of testing is slow, dull and entirely invisible, and it is most of the difference between a plan that works and one that only looks right.

Getting the session onto your bike#

A plan in a chat window still has to reach your head unit. Stride converts each workout into each platform's own format and sends it to Garmin, Wahoo, Hammerhead, COROS, Zwift and Rouvy, scheduled on the right day with the right targets.

Every one of those has its quirks. One will accept a workout, report success, and quietly drop every target if a single reference value is missing, so you start the session and find nothing to aim at. You only find that out by riding it on real devices.

Native apps, not a chat window#

Alongside the web app, Stride has native apps for iPhone and Android. They do the things a chat window can't:

• Indoor training. Ride structured workouts on your smart trainer straight from your phone, with ERG control over Bluetooth, live metrics and automatic upload when you finish.
• Health data. Sleep, HRV and resting heart rate come in from Apple Health on iPhone and Health Connect on Android.
• At a glance. Widgets, Live Activities during a workout, and a notification when your insight is ready.

It runs without being asked#

A home-built coach waits for you to ask. Stride doesn't. Finish a ride and an insight is written from that session plus your recent training, recovery and load. If you are overreached or have missed sessions, it adjusts the coming days and updates your calendar and devices. Weekly and monthly reports arrive on their own.

Behind that sits the engineering nobody sees: testing models against each other, keeping years of history within reach of the AI without the cost running away, and fixing things when a provider changes something overnight.

Built with athletes at the top level#

We work with coaches and professional riders at the highest level of the sport. A good deal of Stride exists because they needed it, and it has been tested against their training rather than a hypothetical athlete's.

So, be our guest#

If you enjoy building things, build it. You will learn a great deal about your own training, and we mean that. If you would rather spend the time training, that is what Stride is for.

And if you like working with Claude, you can have both. Connect it to Stride and let it work from everything we have already collected, cleaned and calculated.