Claim
Willingness to experiment with AI at the edges of what seems possible is the mechanism by which some practitioners discover capabilities months before they become general knowledge. The advantage exists because the AI frontier is jagged and confusion is equal. The benefit goes to those who probe rather than those who wait.
Mechanism
AI capability does not arrive packaged as a public announcement. It sits latent in the model, accessible to anyone who tries the right prompt or framing, discoverable only through deliberate experimentation. In the period before a capability is announced, the field is in shared confusion: nobody knows what is possible. That shared confusion is a flat playing field. Agency is the only differentiator on it. Those who test find what is there; those who wait discover it second-hand.
Conditions
Holds when: AI capabilities are genuinely novel and undocumented, which describes most of the time given how fast the frontier shifts. The advantage is greatest in the window between a model release and the wider documentation of its capabilities.
Fails when: the capability is already well-documented, demonstrated, and taught. At that point, agency no longer gives an information edge.
Evidence
Mollick frames the overhang as a moment where potential outpaces use, and identifies the mechanism that closes the gap:
"a willingness to test the boundaries of what's possible when everybody is equally confused about what AI can do"
The claim is that this willingness, not technical depth or tool access, is the variable that separates early discoverers from later adopters.
Signals
- You find AI can do something months before anyone else in your network mentions it.
- Your experimentation log contains failures you learned from, not just prompts that worked.
- You treat AI refusals or failures as information about the frontier, not as final verdicts.
Counter-evidence
Systematic capability research from labs and third-party benchmarkers partially reduces the agency advantage by surfacing capabilities formally. As evals mature and expand in breadth, independent discovery matters less. Organized communities also aggregate discoveries quickly, shortening the individual advantage window.
Cross-references
- When AI makes production cheap, selection and taste become the scarce resource: as production cost collapses, the bottleneck shifts to selection; agency determines what gets tried, which is now the scarce variable.
- Agentic AI shifts the core skill from prompt phrasing to delegation clarity; instructing AI now resembles instructing a capable but unfamiliar colleague: experimentation precedes delegation. You cannot delegate what you have not discovered.