Bug Blindness: Why Builders and Fans Can't See Their Software's Flaws

Dan Luu argues that most people, including many programmers, are effectively blind to the bugs and quality problems around them—and that this blindness can be cured with practice. Once you learn to notice defects, he says, they become impossible to unsee, which is why executives have repeatedly asked him to evaluate products when they want an honest read on whether something actually works. His findings range from mild annoyances to cases so broken that a normal user could never complete the intended task, even though internal chatter insists the product is great.

The puzzle he keeps returning to is the gap between that internal confidence and the real user experience. He walks through examples: major search engines (Google, Bing, and Kagi) returning pages of SEO spam and even scams, yet Kagi fans insisting their spam-filled results were good; Volvo owners dismissing reliability data because their own car hasn’t broken down; and Blackboard, one of the most detested products in education, being described as beloved by an employee who genuinely believed it. The pattern is that enthusiasm and proximity to a product breed a kind of motivated denial about its faults.

The sharpest example is Discourse, whose staff believed the forum software had excellent web performance. Luu found code that deliberately slowed real page loads to game metrics like Largest Contentful Paint—optimization that crossed into outright cheating, harming users while flattering the benchmark. His broader point for engineers is that if a product feels badly broken when you use it normally, it almost certainly is, and modern LLMs can now simulate ordinary users to confirm that such failures reproduce across many scenarios rather than being rare corner cases.