How do human strengths help people exploit AI capabilities?

Ethan Mollick's claim is that debate over future AI risk distracts from a simpler fact: "AI, right now, is already incredibly capable," and "the current capabilities of existing models are barely being used, and are often not even well understood." He calls this gap "the capability overhang, the gap between what these models can do and what almost anyone is doing with them," and argues it "is an opportunity because most people don't bring their own advantages to AI, and those who do get much more out of it." He illustrates the overhang with demonstrations — an AI rebuilding Zork as a 3D game, reconstructing Umberto Eco's library from videos and photographs, and producing a book trailer using Blender, voice, music, and a video generator without being told how — each of which "would have taken weeks of human work involving researchers, coders, and designers."

Mollick's reasoning is that because "you are not trying to compete with AI in producing outputs, that is a losing game," the exploitable advantage is specific human traits that AI cannot supply on its own. He names four: deep knowledge (expertise that lets a specialist "quickly and accurately make decisions" and judge where "the Jagged Frontier" of AI competence actually lies), wide knowledge (familiarity with enough fields and vocabularies — design thinking, Bayesian reasoning, Rogerian therapy — to know what to ask the AI for, since "AI tends not to volunteer any of these patterns unless you know to ask"), taste (the ability to select among abundant AI output now that "making is fast and cheap" and "the scarce resource is your ability to select among stuff using your own taste"), and agency ("a willingness to test the boundaries of what's possible when everybody is equally confused about what AI can do"). He also cites unspecified "recent work from Anthropic" for the claim that "expertise also shapes the quality of what AI gives back. Experts not only get better work out of AI, they get more work out of it."

That expertise claim sits in tension with Does AI assistance help less experienced workers most?, a study of 5,172 customer-support agents where gains concentrated among the least experienced and the most experienced saw small quality declines. Mollick's claim and that finding describe different tasks — open-ended creative and strategic work he demonstrates personally, versus live chat suggestions in a narrow support role — so the two are not a direct contradiction, but they point in opposite directions on who benefits most from AI assistance, and Mollick's version rests on an Anthropic finding the excerpt does not name or quote directly.

The excerpt is a personal essay built on the author's own demonstrations and anecdote, not a study: the Zork, Eco, and trailer examples show what one well-informed user with deep domain knowledge of Eco, Zork, and his own book accomplished, not a measured comparison against other users or against the same tasks done without his specific expertise. The "recent work from Anthropic" behind the expertise-quality claim is referenced, not cited or quoted, so its scope and method are unknown. The defensible implication is narrow: for someone who already has domain expertise, cross-field familiarity, a clear point of view, and willingness to experiment, current frontier models leave substantial unclaimed capability on the table; whether these four traits generalize as the main predictors of AI benefit across a broader population remains Mollick's assertion rather than a demonstrated result here.