Study Your Industry's History

Most great art is derived from other art

Talent agent and dealmaker Michael Ovitz states the claim directly: "most great art is derived from other art."¹ His evidence runs across two industries. In film, the most acclaimed directors are also the field's most encyclopedic historians: Spielberg, Scorsese, and Kubrick knew more about film history and craft than almost anyone around them, in contrast with film students whose knowledge, in Ovitz's account, effectively "stops at 1990" and cannot place directors like David Lean, Frank Capra, or Michael Curtiz. Spielberg rewatched Lawrence of Arabia annually while he was learning to direct.

In painting, the same pattern holds. A contemporary artist Ovitz discussed at length could be traced through a direct lineage back to Lucian Freud, and knowing that whole chain was inseparable from understanding the work itself. Picasso bought roughly fifty African masks for a few francs at Left Bank markets, and his early faces come directly from those masks, appropriation functioning as a working method rather than a shortcut.

The founder version

Eric Glyman's own practice, modeled explicitly on Ovitz's example, predates this particular account and reinforces it from an entirely different field. Glyman read the complete history of the BankAmericard, reasoning that "how can you create something different, new, and better if you don't understand the shape of where you sit?...it allows you to get to insights faster the more you know."² The founder version of the claim is the same mechanism stated as a research method rather than a creative instinct: read your industry's full history before trying to improve it, because doing so surfaces insight faster than working from the present state alone.

Why it matters

A field's canon functions as compressed search. It shows what has already been tried, what a given move actually signals within the field, and by elimination, where the unclaimed territory sits. It also feeds the kind of contextual judgment described in taste-as-moat, since a filter for what counts as derivative versus genuinely new only works once it has been trained on the full lineage. The claim also compounds with the broader observation that AI compresses domain learning curves: modern tools can hand a newcomer a field's working vocabulary in a matter of hours, but the judgment to tell derivative work from genuinely new work still depends on knowing the actual lineage, which no amount of vocabulary substitutes for.