Two cards from one operator, one source. Both from Ethan Mollick's "The Overhang" (September 18). The library captured the selection claim from that essay on September 19. These two claims were missed then. They are the vocabulary and agency dimensions of the same overhang argument.
What you know determines what AI gives you
Mollick's core observation in Wide knowledge beats narrow expertise when prompting AI, because AI only surfaces patterns you know to ask for is that AI encodes patterns from many domains but only surfaces them when the prompt frames the right ask. Wide cross-domain knowledge provides the vocabulary to frame those asks. Narrow specialists, even deep ones, cannot request patterns they cannot name.
The mechanism is not about AI capability but about prompt framing. A person who knows design critique vocabulary can ask for a specific fix. A person who lacks it cannot ask and therefore cannot receive. Mollick is direct on this:
"AI tends not to volunteer any of these patterns unless you know to ask"
And the flip side:
"If you do know the right terms, asking for changes is easy."
This claim extends two existing Mollick cards. When AI makes production cheap, selection and taste become the scarce resource established that the bottleneck has shifted from production to selection. Wide vocabulary affects selection as much as it affects prompting. Agentic AI inverts the AI-access argument: it amplifies domain experts rather than equalizing non-experts holds that domain expertise increasingly determines what AI can help you do. The vocabulary claim is a concrete mechanism for why.
- Ethan Mollick, Wide knowledge beats narrow expertise when prompting AI, because AI only surfaces patterns you know to ask for. Wide cross-domain knowledge provides the vocabulary to ask AI for patterns it would not volunteer unprompted, making the prompter's knowledge base the binding constraint on AI output quality.
Agency is the only edge on a flat frontier
Agency to test AI limits is the only way to discover capabilities before they are publicly announced addresses a different bottleneck: discovery before documentation. AI capabilities sit latent in models, accessible to anyone who tries the right framing. The window before capabilities are publicly documented is a period of shared confusion. Nobody knows what is possible. That flat uncertainty means agency is the differentiator:
"a willingness to test the boundaries of what's possible when everybody is equally confused about what AI can do"
The word "everybody" carries the weight. The confusion is not specialist confusion or amateur confusion. It is uniform. Those who experiment in that window find capabilities; those who wait learn from others who did. The overhang is not just about unused production capacity. It is also about undiscovered capability that agency closes before announcements make it common knowledge.
- Ethan Mollick, Agency to test AI limits is the only way to discover capabilities before they are publicly announced. Willingness to experiment at the edges of what AI seems able to do is the mechanism by which practitioners discover capabilities before they become general knowledge, because shared confusion makes agency the only differentiating variable.