The Discovery Problem: Why AI Still Makes Users Guess What to Ask

Summary

The article argues that the biggest bottleneck to wider AI adoption is not necessarily capability, but discovery: users often do not know what to ask for. When capabilities are hidden behind a blank text box, people cannot see what the system might do until they invent a prompt and run it. Templates reduce that burden by providing ready-made starting points, but they may not match a user's actual work or interests. Personal context can make suggestions more relevant, yet it still does not expose the full range of possibilities. Using Alan Kay's ant-at-the-bottom-of-the-Grand-Canyon metaphor, the article compares ordinary users with people fluent in agents and tool use: both may have access to the same system, but the experienced user can see many automations and delegations that the other person has never considered. The author argues that too much capability discovery currently falls on users who lack the vocabulary to request it. A more mature interface would gradually and contextually reveal useful options based on the user's actual work. The article concludes that AI systems still have not learned to show users the broader "sky" of what they can do.