Runway CEO connects video generation to advertising agents and future robotics

Runway co-CEO Cristóbal Valenzuela sees the company's video-model work extending beyond generated clips into advertising agents, interactive software and robotics. In an interview with The DeepView, he describes those areas as connected by models that learn how the world changes over time. The interview also makes clear that some of the proposed applications remain early or speculative.

The DeepView conversation, conducted by Nat Rubio-Licht around Runway's summit, combines released product discussion with Valenzuela's forecasts. Keeping those categories apart matters: an advertising workflow discussed as a current product is different from an estimate of when world-model policies might reach live robot deployments.

Advertising gives the agent an observable workflow

Valenzuela describes Runway Ads as an agent that can create, edit, publish and track advertising performance. The proposed loop gives the system more to do than produce a video file: it connects creation with distribution and feedback. His interview frames that as an initial step toward broader agent interactions.

The previous Runway Ads announcement described generating image and video variants from a brand kit, publishing approved versions to Meta, Google and TikTok, then using platform results for another batch. That earlier product description provides concrete context for the advertising example in this interview. It does not establish that every campaign improves, or that generating more variants automatically produces a better business result.

Valenzuela also discusses creative agents as collaborators in longer fictional or interactive experiences. Those comments outline a direction for the technology rather than a fully specified product with established access, pricing or performance. The interview does not give enough detail to treat all of those ideas as available features.

Generalization depends on engineering and data preparation

The CEO argues that video models can learn capabilities useful across creative production and robotics. In his account, predicting how a scene changes is relevant to more than filmmaking. That is his explanation of Runway's strategy, not independent proof that a video model can control any robot or generalize to every physical environment.

He points to practical compute constraints as one reason progress can be slower than expected. He also emphasizes the work of collecting, cleaning, processing and annotating diverse data. The discussion therefore concerns more than increasing model size: usable training material and the infrastructure to process it remain part of the work.

The DeepView report describes Praxis-1 as an open-weight world-action model that can adapt to a new robot embodiment with lightweight fine-tuning. That description comes from the interview report; Franklin has not run the model or reproduced its adaptation results. A claim about adapting a model is also distinct from demonstrating safe, reliable operation in a production setting.

The robotics timeline is a forecast, not a deployment result

Valenzuela estimates that fully live production deployments could be 12 to 18 months away. He discusses remaining engineering and performance-optimization challenges. The estimate should not be read as a guaranteed shipping date, a customer deployment already achieved or evidence that the remaining challenges are resolved.

His discussion of Solaris and interfaces generated through models extends the same ambition into software experiences. When the conversation reaches generating an entire operating system, he explicitly calls the idea early and speculative. That qualification is more informative than presenting the possibility as another finished launch.

The interview's clearest contribution is to explain how Runway's leadership connects several bets: advertising offers a constrained action-and-feedback loop, creative experiences broaden interaction, and robotics brings the additional demands of physical deployment. Whether one foundation successfully serves all three will need product results and evaluations beyond a CEO's strategic account.