Google tests Agent Mode on Flow

Google looks like it is turning Flow from a prompt box into a production assistant. New app references point to an unreleased Agent Mode inside Flow — Google’s AI filmmaking tool — with a toggle that would let creators switch between normal prompting and a more autonomous helper. (testingcatalog.com) ### What is Flow again? Flow is Google’s filmmaking app built around three model families: Veo for video, Imagen for images, and Gemini for language and prompting. The whole pitch has been “make cinematic clips and scenes in one place,” with controls for characters, locations, objects, and shot style rather than just one-off text-to-video generations. (blo([testingcatalog.com)ek? A teardown of recent Flow builds surfaced signs of an Agent Mode that is not live yet but is clearly being wired into the interface. The notable part is where it sits — directly on the prompt bar — which suggests Google wants this to feel like a first-class way to work, not a buried experimental setting. (testingcatalog.com)de actually do? Basically, it looks like a handoff mechanism. Instead of typing every shot, revision, and sequencing step yourself, you would be able to ask for an outcome and let the system manage more of the project assembly. The reporting points to creators choosing between conventional prompting and heavier orchestration by an automated helper. (testingcatalog.com) ### Why is that a bigger deal than another model update? Because the bottleneck in AI video is no longer just “can the model make a pretty clip?” It is “can I keep characters consistent, manage assets, revise scenes, and move from idea to finished sequence without juggling five tools?” Flow already tries to solve part of that by keeping story ingredients in one workspace. An agent layer would push further — from generating media to coordinating the work itself. (blog.google) ### Has Google been moving this way elsewhere? Yes — and that is what makes the Flow leak believable. Separate recent reporting has pointed to a broader Agent Mode push in Gemini, including workflows, scheduled actions, and task-oriented skills, with Google I/O on May 19–20 framed as the obvious place for that story to come together. So Flow’s agent features look less like a one-off experiment and more like part of a wider product direction. (testingcatalog.com) ### Where does Veo fit into this? Flow is the wrapper; Veo is the video engine underneath. Google launched Flow around its Veo stack at I/O 2025, then kept adding more editing and control features, including later updates tied to Veo 3.1 and integrated image workflows inside Flow. That matters because Agent Mode would sit on top of a system that already has enough controls to automate something meaningful. (techcrunch.com) ### So is this available now? No — at least not publicly. What surfaced is scaffolding in app builds, not a launched feature. The catch is that interface clues tell you intent, not shipping dates. Google could announce it at I/O, test it quietly, or change the design before release. (testingcatalog.com)I model and an AI coworker. A model gives you clips. An agent could help manage scenes, revisions, and continuity across a whole project — more like a producer’s assistant sitting beside the timeline than a vending machine for shots. That is where real workflow lock-in starts. ### Bot(testingcatalog.com)it wants Flow to help run the production process. (testingcatalog.com)