The AI feature everyone ships and almost nobody uses correctly

Working inside AI product teams for the past few years, I've watched the same feature get shipped, celebrated in a launch post, and then quietly used by a fraction of the users it was built for: the "summarize this" button, bolted onto everything from email to meeting notes to long documents.

The usage data tends to tell a consistent story — a spike in the first week from curiosity, then a steep drop-off, then a small stable core of users who actually kept using it. What separates that stable core from everyone else isn't the feature itself, it's a habit: they use summaries as a triage tool to decide what deserves a real read, not as a replacement for reading anything that matters.

The people who tried summarization as a full replacement for reading reported the worst outcomes — missed nuance, wrong conclusions, decisions made on a compressed version of something that needed the detail. The people who used it as a filter, to decide which three of twenty documents deserved actual attention, reported it as one of the most useful tools they'd adopted all year.

Same feature, same technology, opposite outcomes, depending entirely on whether it's used to decide what to read or used instead of reading. That distinction rarely makes it into the launch post.