Netflix bought a filmmaker, not a tool

Netflix just acquired InterPositive, an AI filmmaking tools startup founded by Ben Affleck. The deal landed quietly, with undisclosed terms and a 16-person team folding into the streaming giant. Most coverage focused on the celebrity angle or the labor implications. But the real story is simpler and more interesting: Netflix didn't buy a technology. They bought a filmmaker who happened to build technology. That distinction matters more than it might seem.

What InterPositive actually does

InterPositive is not a text-to-video generator. It doesn't compete with Sora or Veo. Affleck was explicit about this: "It's not about text-prompting or generating something from nothing." Instead, the tools work within an existing production. A filmmaker shoots footage as usual, and InterPositive builds a custom AI model from that production's own dailies. That model can then help with postproduction tasks like relighting shots, removing stunt wires, reframing compositions, replacing backgrounds, and correcting continuity issues. The key detail is that the model is trained on the production's own footage, not on a generic internet-scale dataset. It speaks the language of cinematographers and directors because it was built by someone who actually is one.

The filmmaker built the workflow, not the other way around

Affleck started InterPositive in 2022, operating it in stealth under the corporate entity Fin Bone LLC. He and a small team filmed a proprietary dataset on a controlled soundstage, deliberately replicating the full conditions of a real production. The goal was to capture what actually happens on set, including the unpredictable parts, and encode that into a system filmmakers would recognize. This is a fundamentally different approach from most AI startups. The typical pattern is: take a powerful model, wrap it in a generic interface, and hope users figure out how to make it useful. InterPositive flipped this. Affleck started with the workflow, the specific frustrations and constraints of postproduction, and built the AI to serve it. His first model was trained to understand what he called "visual logic and editorial consistency." It preserves cinematic rules under real-world conditions. It includes guardrails to protect creative intent. These are not features you arrive at by thinking about AI capabilities. They're features you arrive at by thinking about what filmmakers actually need.

Domain expertise is the real moat

This is the pattern worth paying attention to. The most defensible AI companies are not the ones with the best models. They're the ones founded by people who deeply understand the problem they're solving. InterPositive works because Affleck has spent decades on film sets. He knows what breaks in postproduction. He knows the vocabulary directors use. He knows which creative decisions need to stay in human hands and which tedious tasks are safe to automate. That knowledge informed every design choice, from the training data to the product constraints. Compare this to the average AI startup. Most of them start with a model and go looking for a problem. They wrap GPT or Stable Diffusion in a thin interface and call it a product. The industry has a name for this: GPT wrappers. And GPT wrappers are commodities. The moment the underlying model improves or a competitor builds the same wrapper, the business evaporates. InterPositive is not a wrapper. It's a filmmaker's workflow encoded into software. The AI is the engine, but the domain knowledge is the chassis that makes it drive somewhere useful.

Why Netflix bought this specifically

Netflix doesn't need AI technology in the abstract. They have an enormous engineering organization and the resources to build or license any model they want. What they can't easily build internally is the deep production intuition that shaped InterPositive's tools. Elizabeth Stone, Netflix's Chief Product and Technology Officer, made this point directly: most generative AI video platforms "don't operate from the perspective of a filmmaker." Netflix wasn't buying compute or model weights. They were buying a product shaped by someone who knows what it feels like to lose a shot, to fight with lighting in post, to wish you could reframe without reshooting. The deal also came just days after Netflix walked away from its bid to acquire Warner Bros. Discovery's studios and streaming businesses. The timing suggests a deliberate pivot: rather than acquiring massive content libraries, invest in tools that make production itself more capable. Distribution plus production infrastructure is a more durable advantage than distribution plus catalog.

The broader pattern

This isn't unique to filmmaking. Across every industry where AI is being applied, the same dynamic is playing out. The companies that win are the ones where the founders brought deep domain expertise to the table, not just technical skill. In developer tools, the most effective coding assistants aren't built by the teams with the largest models. They're built by people who understand how developers actually work, how they navigate codebases, how they debug, what context they need. Cursor and Anthropic's Claude Code have gained traction not because of raw model performance alone, but because their products reflect a genuine understanding of programming workflows. In healthcare, the AI diagnostic tools gaining clinical adoption are the ones built by physicians who understand the messy realities of patient care, not by ML engineers optimizing for benchmark accuracy. The pattern repeats: when intelligence becomes commoditized, the differentiator is taste. Knowing what to build, what to constrain, what to leave in human hands. These are judgment calls that require lived experience in the domain.

What this means for AI startups

If you're building an AI company, the InterPositive acquisition is a signal worth reading carefully. Your moat is not the model. Models are improving rapidly, costs are falling, and access is democratizing. Anything you can do with today's model, someone else will be able to do with tomorrow's model for less. Your moat is your understanding of the workflow you're automating. The specific pain points, the edge cases, the unspoken conventions, the judgment calls that practitioners make instinctively but can't easily articulate. If you can encode that understanding into a product, you have something that a better model alone cannot replicate. Affleck built InterPositive's training data on a soundstage, not scraped from the internet. He built constraints into the system to protect creative intent, not because it was technically interesting, but because he knew from experience that filmmakers would reject any tool that overrode their judgment. These decisions came from domain expertise, and they're what made the product worth acquiring.

The scarcest input

In 2022, the scarce resource in AI was compute. In 2023, it was data. By 2025, both had become increasingly abundant. Models got cheaper, datasets got larger, and the barriers to building basic AI applications dropped to nearly zero. What remains scarce is the ability to look at a powerful general-purpose technology and know exactly how it should be shaped for a specific context. That's not a technical skill. It's the accumulated judgment of someone who has spent years, sometimes decades, inside the domain. Netflix understood this. They didn't buy InterPositive for the AI. They bought it for the filmmaker who knew what the AI should do. In the age of commoditized intelligence, the scarcest input is taste.

References