A rewrite should be invisible
Sonos rebuilt its app, shipped it missing features people used every day, and now says fixing it will cost $20 to $30 million. A rewrite should change nothing a customer can see on the first day.
Sonos shipped a rebuilt app on May 7. It arrived without things people used every day, and the backlash hasn’t stopped since. On July 25, CEO Patrick Spence apologized and published a timeline: music library setup, browse and search through August, alarm reliability in September, and editing playlists and the queue in September and October. On August 7, Sonos put the cost of fixing it at $20 to $30 million and delayed two products planned for later this year.
Spence’s case for the rewrite was reasonable: about 20 years of accumulated performance and reliability problems, and a new modular platform to build on. I believe him. Old code holds a company back, and someone has to pay to replace it.
The mistake was letting customers see the bill.
Underneath is not their problem
A rewrite should change nothing customers can see on its first day, because the people who depend on a product don’t care what’s underneath. They care that the alarm goes off and that they can find the album. Everything the new foundation makes possible can arrive later, once the old jobs work the same as before.
That means the first release of a rewrite is meant to be boring. The bar is parity with the old app, and the goal is that nobody notices. Then, one release at a time, you spend the new foundation on things people can feel.
Sonos did it in the other order. It shipped the new foundation with the old jobs half done, and asked customers to wait for parity.
Everyone in the house
What makes this worse for Sonos is who uses it. A speaker in a kitchen isn’t one person’s device. Most of my work on Android has been a screen a whole household shares, and shared devices have a particular failure mode. When something breaks, it breaks for everyone in the room at once, including the people who never chose the product and don’t know how it works. One person updates the app. The whole house finds out when the alarm doesn’t go off.
So a shared device needs a stricter rule than a personal one. Parity first, then everything else, and a way back to the old version until parity is proven.
What it costs to do it right
Running two apps side by side while the new one catches up is expensive, and it slows the new platform down. It’s also cheaper than $20 to $30 million and two delayed launches. The old jobs are the reason anyone owns the product, so they ship first.