Sixteen People

In March, Netflix acquired InterPositive, an artificial intelligence startup founded by Ben Affleck, for up to 600 million dollars. The company brought a team of sixteen people.

Whatever else that transaction was, it was a price. Roughly 37 million dollars a head for a capability the buyer had decided it could not assemble fast enough internally, in an industry that two years earlier had shut down production over the same technology.

The disclosure is the news, not the deal

On its second-quarter earnings call Netflix said generative AI had been used in approximately 300 titles in 2026, mainly in post-production. Named examples include Glory, Brasil 70: A Saga do Tri and The American Experiment, where the tools built crowds and historical battle sequences, with reported cost and time roughly halved on those specific sequences.

A studio does not put that number in front of investors casually. Three hundred is large enough to be a policy rather than a set of experiments, and disclosing it on an earnings call converts it from a production detail into a statement about operating model. The company is telling the market that its cost base has structurally changed.

The qualifier is doing real work and deserves to survive summary. Post-production. The savings attach to shots, not to films: crowd extension, historical recreation, cleanup, format conversion. Principal photography, development and the judgment about what should be made are not where these tools are operating.

Two deals, two different theories

In June, Google DeepMind invested 75 million dollars in A24 to develop filmmaking tools. On the reported terms the arrangement does not give Google access to A24's content library or its data. A24 subsequently had to defend the partnership publicly, which is itself a datapoint about where the industry's sensitivities now sit.

Set the two transactions beside each other and they describe different theories of what is scarce. Netflix bought a team, on the view that the constraint is people who can build production tooling. Google bought proximity to a working studio, on the view that the constraint is understanding what film production actually requires, having discovered that general video generation is not the same thing as a usable workflow.

The A24 terms matter beyond that deal. Most creative-sector AI arrangements to date have exchanged preferential tooling for training material. An explicit, reported separation of the two establishes that a studio can decline to trade its library and still get the partnership, which is the sort of precedent that shapes every negotiation behind it.

What has not been settled

Not consent. The talent side has converged on a three-part test, consent, controls and compensation, and it is a reasonable frame that remains unevenly implemented. The exposure is largest not at the studios, which have negotiated agreements and legal departments, but everywhere downstream: the advertising, corporate and independent production layers where releases signed years ago say nothing about synthetic derivation because the concept did not exist.

Not disclosure. The industry's working posture has been described as don't ask, don't tell, and it is unstable for a straightforward reason: it depends on audiences not caring, and the only evidence anyone has about whether audiences care is that the people making the decisions would rather not find out.

Not the labour question, though it has changed shape. The 2023 dispute was about whether the tools would be used. The live question is which specific crafts are compressed, and the answer emerging from Netflix's own description is the technical middle of post-production rather than performance or direction. That is a narrower displacement than was feared and a real one for the people in it.

The pressure underneath

None of this is primarily about capability. Short-form platforms now command more viewing time in the 13 to 54 bracket than the major studios' combined output, and the studios' response has been to cut existing content into vertical formats to compete for that attention.

That volume is not achievable at traditional post-production economics. The demand is not for one finished asset but for that asset in many aspect ratios, lengths and openings, refreshed constantly. Once the requirement is stated that way the tooling follows automatically, and it becomes clear that the technology did not create the pressure. Distribution did, and the technology arrived in time to be the answer.

Which suggests the durable question is not whether AI belongs in production. It is whether an industry reorganising itself around the attention economics of vertical video ends up making anything it would have chosen to make. Three hundred titles is an operating statistic. It is not yet an answer to that.

Sources

• Variety and Engadget on Netflix's second-quarter earnings disclosure. Source for generative AI in roughly 300 titles in 2026, primarily post-production, the named titles, the crowd and battle work, and the reported halving of cost and time on those sequences.
• Reporting on Netflix's March 2026 acquisition of InterPositive, founded by Ben Affleck, for up to 600 million dollars, bringing a sixteen-person team, with Affleck taking a senior advisory role on filmmaker-facing generative tools. The per-head figure in the opening is this publication's arithmetic on those reported numbers.
• TechCrunch, Variety, Google and Deadline on the DeepMind and A24 partnership, June 2026. Source for the 75 million dollar investment, the reported exclusion of library and data access, and A24's public defence of the deal.
• The Economist, issue of 8 to 14 August 2026, for the three Cs framing, the short-form viewing-share comparison and the don't ask, don't tell characterisation. Its account of a ByteDance video model and an associated takedown dispute does not match the public record on the products named and is not used.