AI video generation, editing and analysis could change our world beyond recognition

New AI models will be able to create and analyze long videos, combining editing, clips and sound the way a production team does. John Werner of Forbes writes that tools such as Opus 5.5 and GPT Astra 6 could change not only video production but everyday communication: instead of a call or a meeting, people will send realistic video avatars, while reports, books and menus will be turned into videos. The author says the consequences of this shift are still impossible to calculate.

Video generation is changing communication, work and reading

From podcasts to “agentic video”

In the summer of the year before last, NotebookLM showed how AI could create a conversation between two podcasters on a given topic. To John Werner, the example recalled new versions of the Turing test: it is becoming harder to tell a conversation with a person from an interaction with a non-human system. Now, he argues, video tools will bring a similar shift.

Werner listened to an episode of the AI podcast Daily Brief in which Nathaniel Whittemore discussed criteria for and possible uses of “agentic video”—video created and interpreted by non-human systems. Whittemore highlighted the ability to “watch, or approximately watch,” a video: understand its content, find key moments, and assess both individual parts and the video as a whole.

Models such as Opus 5.5 and GPT Astra 6, Werner suggests, will be able to create long videos with coordinated sound, editing and clips, as if a team of people had worked on them. Sound is especially important here: the author believes it will help new models cross the “uncanny valley.”

Avatars instead of meetings

In the section “Your Digital Twin,” Werner looks at video avatars—realistic animations of users themselves. Instead of answering the door in a bathrobe, making a call or meeting for coffee, someone could send a friend or neighbor an avatar to say something on their behalf. As a result, people may see and hear one another more often through these mediated formats.

Whittemore also discusses business uses for video. For example, a model could prepare a video summarizing the main takeaways from a Zoom meeting. This leads him to ask: if a meeting is not worth turning into a video, why hold it at all—could people not just send avatars instead?

Another scenario is multilingual support. Instead of a bilingual intermediary translating a phone conversation between people who do not understand each other, AI could provide a reasonably accurate approximation of translation—at the level of “three nines,” or 99.9%.

Video instead of reading

Werner develops another of Whittemore’s ideas: people who do not want to read can watch a video. The author believes that is almost certainly the case today. If executives do not read reports, video could help capture their attention. But Werner extends this logic to fiction and nonfiction, policy proposals, menus and other materials that previously existed in written form.

The author suggests that audiences will increasingly prefer video because it is better at holding their attention. He wryly observes that this shift would have Marshall McLuhan “turning in his grave.”

Werner writes that a flood of deepfakes could set off a race to deliver messages: instead of text, people will receive videos featuring a narrator, perhaps someone’s avatar. People will stop reading. That could improve posture and remove barriers to communication, but it would look strange: the person in the video is not actually speaking—they have only explained to the system how to speak and, to some extent, what to say.

The author acknowledges that the consequences cannot yet be fully calculated, but says one thing is clear: vivid video generation will change the world significantly. In Dataist’s view, in these scenarios video becomes a way to create content, deliver messages and replace face-to-face interaction. But the boundary between a person and their digital representation may become less distinct, while the scale of the shift remains unclear.