In 2015, a film called Tangerine premiered at Sundance to critical acclaim. It was shot entirely on the iPhone 5s. Its director, Sean Baker, went on to win the Academy Award for Best Picture in 2025 for his film Anora. In the years between, studios released a steady stream of expensive disappointments shot on the finest cameras ever made. Whatever separates great films from forgettable ones, it is not the camera.
This is something everyone already knows about movies. Better cameras genuinely matter. No director shoots a blockbuster on a flip phone by preference, and the history of film is partly a history of technology opening new possibilities. But camera quality is one input in an equation dominated by other terms: the craft of writing, directing, and editing that brings a thousand moving pieces together into something coherent. A novice with a cutting-edge camera loses to a great director with a terrible one every time. The camera records the film. It does not make it.
Right now, organizations are treating AI as if it is a camera that makes a film on its own. Models are improving fast, and better models will help. But AI model power is only one part of the equation. Everyone has access to the same models, just as every studio has access to the same cameras. What shapes the return is everything wrapped around the model.
The two things wrapped around the model
That everything comes down to two things. The first is the craft: conventional software development experience. This is the accumulated, unglamorous skill of building systems that actually work — how the pieces connect, where they break, what to build and what to buy, how to ship something people can rely on. It is the screenwriting of the operation: the structure underneath everything the audience sees, the discipline that decides how every scene connects and how the pieces set up and pay off. A film with a weak script fails no matter what it was shot on.
The second is the direction: organizational and strategic knowledge. A director decides what story is worth telling and how every role on the set serves it. The organizational equivalent is knowing what your company actually needs, who will use the system and how they really work, and how a new tool fits into the operation that already exists. This knowledge cannot be purchased in a mass-market way, because it is knowledge about your organization itself.
Put differently: the model is the camera, the software is the screenplay, and the organization is the direction. An implementation that gets all three right can produce remarkable results with ordinary technology, the way Tangerine did. An implementation that has only a fancy camera produces expensive footage of nothing in particular.
If your AI investments are disappointing
The implication runs against the current instinct. If your AI investments are disappointing, the fix is probably not a more powerful model, for the same reason a better camera will not save a film with no direction. The organizations seeing real returns are the ones building filmmaking ability: engineering craft and strategic judgment about their own systems and people. That ability compounds, it transfers from project to project, and unlike the technology, it cannot be bought by a competitor the same afternoon.
Everyone is buying better cameras. The returns will go to whoever finds a great director.