Agentic media production: here sooner than predicted

At our Gen AI Conference in June, Daniel Mahlow predicted that media production would become fully agentic. His assessment today is that it went faster than expected. October alone produced four pieces of evidence.

Google released Nano Banana 2.1, a cheaper image model. Canva introduced Craft, its own model, which agents can call directly. Noiz AI showed Helix World, a walkable scene with sound generated from a single image. The furthest step comes from Reka with the preview of Rho-1, a model that processes text, image, video and robot actions in a single network (Reka).

Why Rho-1 is the one to watch

In July we described a production chain here in which an agent hands concept, storyboard, video, voice and editing to specialist models one after another.

That chain is exactly what Rho-1 calls into question, because every handover costs time and each specialist model only ever sees a fragment of the whole.

Our take

For marketing and communications the bottleneck has moved for good — away from production and towards briefing and selection. Anyone who gets ten variants in an hour above all needs a reliable judgement about which of them fits the brand. That judgement is not a model capability. It is a decision someone has to own.

Architecturally there are two bets side by side: orchestrating specialist models on one side, omni-models on the other. Which direction wins cannot be called today, which is why interchangeable building blocks are the safer choice. Build so that swapping the model is a configuration change, not a project.

And then there is checking the output. At ten assets a week a human still does it. At a thousand, nobody does — which means the check has to be designed in before the volume arrives, not after.

Thinking about your own production chain? Daniel is happy to talk it through.