If you’re a tech pro of a certain age, you no doubt have traumatic memories of Clippy, Microsoft’s “Office Assistant” (rendered as a cheerful, anthropomorphic paperclip) that would pop up at inopportune moments to ask if you needed help writing a letter (for example). After a lot of blowback from users, Microsoft eventually removed Clippy from Office. That didn’t stop years of memes and jokes about the assistant. In one season of HBO’s “Silicon Valley,” for example, the show’s startup heroes create “Pipey,” which (like Clippy) is unbelievably intrusive and annoying: Clippy is so toxic, in fact, that when he (she? it?) recently made a brief appearance as a sticker in Microsoft Teams, the company's Slack clone collaboration software, the reaction was immediate—and Clippy was promptly vaporized again. “Clippy has been trying to get his job back since 2001, and his brief appearance on GitHub was another attempt,” Microsoft wrote in a statement to The Verge. “While we appreciate the effort, we have no plans to bring Clippy to Teams.” (Although the Clippy stickers appeared on GitHub, Microsoft's "brand police" were reportedly unhappy about it.) Although Clippy is a punch line, it’s also an excellent cautionary tales for developers (indeed, anyone who works with software). After all, Microsoft didn’t build the world’s most annoying virtual assistant in a vacuum; they used a variety of research methods to create something they thought would work. Kevan Atteberry, who was responsible for Clippy’s work, once told Motherboard about the effort involved:

“We designed about 250 characters, and I had about 15 or 20 of 'em in there. Through working with some social psychologists out of Stanford, we spent six months going through them all, whittling them down with focus groups and stuff like that, and [Clippy] came out to be the number one most trustful and engaging and endearing character of them all. So he became the default.”

Despite all that work, Clippy was still polarizing from the first moment he popped onto folks’ Office screens. According to Atteberry:

“But to be honest, not everybody hates him. I get a dozen pieces of fan mail from people that just loved Clippy. Even some of the responses on Twitter you can see with the pregnant Clippy, there are some people that want to bring him back. And uh... you know.... People hate him. At one point he was annoying hundreds of millions of people a day, which was kind of funny.”

What’s the lesson for developers? You can focus-test a new feature—and even land some positive feedback—only to find the negative response overwhelming once you release it to the broader world. That’s an unavoidable scenario, no matter how good your research or prep work. However, Microsoft compounded its error by not killing Clippy as soon as the tsunami of negative feedback rolled in. This was partially due to old-style release cycles, which prevented companies from easily tweaking customers’ software within days or weeks; everyone was stuck with Clippy until a new version of Office hit store shelves. But the company also made a decision to stick with Clippy through a couple of Office iterations, suggesting that someone internally really, really loved it despite the screams of agony from around the world. Today, developers have the ability to push an update, meaning they can vaporize many bad features quickly. And perhaps they should be merciless with that ability; the last thing you want is an awful icon, widget, or function to become a meme. When in doubt, consider whether you should kill.