Netflix crosses 300 generative-AI productions, and quietly redraws the line on craft

Roughly 300 Netflix programs have used generative AI somewhere in their production process this year, the company disclosed in its second-quarter earnings report on Thursday 16 July 2026, more than double the figure Netflix reported a year ago. The number was tucked into a wider discussion of cost discipline, but it lands at a moment when the rest of Hollywood is still arguing about whether the technology belongs on set at all.

The disclosure is the largest of its kind from a major US streamer and the clearest signal yet that generative AI has moved from a handful of opt-in pilots into the default toolchain for premium television. It also sharpens a question the industry has been ducking: if machine-made imagery is now ordinary, what counts as craft?

From pilots to production line

Co-chief executive Ted Sarandos said on the earnings call that the 300-program figure spans visual-effects work, pre-visualisation, shot-referencing and other behind-the-scenes tasks rather than the on-screen performance of any single actor. The framing matters: Netflix is not claiming it has replaced writers, directors or performers. It is claiming it has rewired the plumbing beneath them.

The 2026 number is roughly double the 2025 tally, when Netflix said about 10 productions had used generative AI end-to-end. That earlier figure was widely cited at the time as evidence the streamer was moving cautiously. The latest disclosure complicates that read. Going from a handful of test cases to a few hundred programmes inside twelve months is not experimentation; it is industrial policy inside one company.

The studio counter-narrative

Not every studio agrees on the pace. Rival platforms and the major agencies have leaned toward narrower deployments, citing concerns about residual obligations, likeness rights and the residual-fee structures that govern most unionised US production. The Screen Actors Guild–American Federation of Television and Radio Artists (SAG-AFTRA) has treated generative AI as a non-negotiable bargaining item since the 2023 strikes, and any deployment that touches a performer's likeness still requires contractual sign-off.

The structural counter-argument from Netflix's side is straightforward: in a format where margins are thin and catalogue churn is high, the technology that wins is the one that lowers the marginal cost of an effect, an environment or a previs pass. Studios that resist on principle accept a cost penalty on every project, and the cost penalty compounds across a slate.

Where the line actually sits

The 300-program figure does not, on its own, resolve the more uncomfortable debate about authorship. A pre-visualisation pass that lets a director see a sequence before a set is built is a long way from a fully machine-generated scene shipped to viewers. The earnings call language is careful to keep that distinction in view.

That distinction is also where the political fight lives. US lawmakers in both parties have proposed disclosure regimes that would require streamers to flag AI-generated or AI-augmented material; the EU AI Act, in force since phased deadlines began in 2025, already imposes transparency obligations on certain AI-generated content distributed in the bloc. Netflix is not currently calling out individual programmes to viewers. Whether it should is now a question for regulators more than for the company's earnings team.

What 300 programmes buys you

The number itself is more meaningful as a direction of travel than as a benchmark. Two years ago, the use case was a single VFX shot in a single show. Three hundred programmes suggests that, inside Netflix's production stack, AI tooling has become as ordinary as cloud rendering or remote collaboration. The studio that reaches that point first does not merely save money; it changes the reference point that every other studio is measured against at the next investor day.

The contest now is over who sets the disclosure floor, who pays the residual, and whether a viewer scrolling the home screen has any way to tell which titles were touched by the technology and which were not. The earnings report has, for the moment, settled only the first of those three.

This piece frames the earnings disclosure as an industrial milestone rather than a creative verdict; Monexus will revisit the disclosure question when US federal labelling rules are formally proposed.