On the front page of Sunday’s New York Times travel section, Stephanie Rosenbloom tells a story about the first time she tried to order a Uber ride through the company’s Apple Watch app. The app seemed to be frozen — until it wasn’t.
“Suddenly,” she writes, “it was ordering me car after car in what seemed like a digital-age version of the ‘I Love Lucy’ candy factory episode.”
Welcome to the world of first-generation third-party apps, the ones giving the Apple Watch a bad name. I’m not the only early user who’s complained about how slow and buggy they seemed.
Now, thanks to Marco Arment, we know why.
Arment is a rock star among app developers. He created Tumblr, Instapaper and what may be the best podcast app on the iPhone: Overcast.
But even he needed two tries to get it right. The first version of Overcast for the watch was a disaster, and he thinks he knows why. It’s not the tiny screen that is the challenge, he explains in a blog post that’s getting a lot of attention this weekend, it’s the load time.
Every time the interface loads or changes, the Watch and iPhone communicate round-trip over Bluetooth. Whether due to wireless flakiness, 1.0 OS bugs, or (most likely) both, WatchKit is frustratingly unreliable. Apps or glances will sometimes just spin forever instead of loading, and even when everything’s working perfectly, apps still take so long to load and navigate that the watch’s screen often turns off before you’ve accomplished anything.
To simplify the code and reduce the workload on WatchKit’s fragile data connection, Arment rebuilt the app from the ground up, re-structuring it around the screen that gets used the most: Now Playing. As he explains,
Trying to match the structure of the iOS app was a mistake. For most types of apps, the Apple Watch today is best thought of not as a platform to port your app to, but a simple remote control or viewport into your iPhone app.
My initial app was easier to conceptualize and learn, and it closely matched the iOS app. But it just wasn’t very good in practice, and wasn’t usually better than taking out my phone.
Apple, whose Podcast app pales by comparison, could learn a lot from Arment. So could Uber.
Follow Philip Elmer-DeWitt on Twitter at @philiped. Read his Apple (AAPL) coverage at fortune.com/ped or subscribe via his RSS feed.