Netflix says it used generative AI in roughly three hundred pieces of content released this year. Not three hundred internal tests, not three hundred slides in a research deck. Three hundred titles that went into the catalogue.
The company pointed to three productions to make the adoption concrete: the Indian title “Glory”, the Brazilian “Brasil 70: A Saga do Tri” and the American project “The American Experiment”. In those productions, according to Netflix, generative AI was used at an operational level, to create and manage sequences the company itself describes as highly complex. Earlier, the service had confirmed that AI was used to build a visual effects scene in “L’eternauta”. Read together, those are not prototypes that escaped a lab. They are a tool working inside production and post-production and staying there.
Now look at the venue. The figure was published in a letter to investors, alongside financial results. Netflix tells shareholders that the use of generative AI by its creative partners is growing rapidly, and that these tools deliver higher quality results, faster and at lower cost than traditional methods. That is the company describing its own work, in a document written for the readers who put a price on it.
Higher quality, faster, cheaper. Three promises in a single line, and all three are stated by the party selling the outcome. A claim made in a shareholder letter is a claim: on its own it does not establish the result it describes, and it deserves the same reading as any other line in a quarterly document, as something the author benefits from you believing.
Then there is the sentence that does the most work. In some cases, Netflix says, without generative AI the productions would have had to give up fundamental shots and sequences. That is an admission offered as a credit, and as rhetoric it lands: the tool walks on stage as the alternative to a sequence that would not have been shot at all. It is also the moment where the company’s own account of the value becomes clearest. The benefit being described is that something exists rather than that something is better.
It is worth being equally precise about what this does not prove. Nothing here says that entire works were generated. Nothing here documents a single person losing a job. These are sequences inside productions made by people, and the published claim stops exactly there. Treating a risk as if it had already been realised is the fastest way to lose an argument that is otherwise easy to make.
So here is the argument. A number like three hundred is not a technical milestone, it is an organisational one. Roughly three hundred titles carrying declared use mean a decision taken, repeated and turned into routine, then reported to the readership that cares most about unit cost. Note what is being counted: pieces of content in which the use is declared, not a tally of how many times a model was put to work inside each one. Read as adoption, the figure works perfectly. Read as anything else, it stops working immediately. It says nothing about whether one single title among them came out better, because that is not the quantity being counted. Adoption and craft are two different measures, and the first does not answer for the second.
If the saving Netflix describes is real, the part worth pressing on is who ends up holding it: the production that would otherwise have dropped the sequence, the people who would have built that sequence by hand, or the results line that announced the number. Roughly three hundred titles show how far the choice has spread. They do not show it was the right choice even once.
Raffaele Di Marzio
All my “insane” books on cybersecurity and governance are here 👇 https://cyberium.limited/bookshelf.html