Mosaic's Motion agent takes on editing and video understanding
The agent cuts, motion-designs and grades video from a brief, then hands back a project you can finish in Premiere, After Effects or DaVinci Resolve.

Key takeaways
- Mosaic published a research release describing Motion as a general video agent that plans, cuts, motion-designs and grades video from a brief or a reference video.
- Motion connects to Google Drive, Frame.io, Notion and Slack, can run on a schedule, and exports projects to Premiere Pro, After Effects and DaVinci Resolve.
- It is callable over an API and an MCP server, so agents such as Claude and Codex can drive it.
- Motion first launched on June 18, 2026; the October 20 post is Mosaic's framing of it as a general video agent spanning editing, creation and understanding.
Mosaic has published a research release positioning Motion, its video agent, as a general one: a single agent for editing, motion design and video understanding. Motion takes a brief, brand assets or a reference video, plans the shots and produces a finished cut, then hands back something you can keep editing.
The agent works two ways. With footage you already have, Mosaic says Motion can cut a multi-camera podcast, find the strongest moments and reframe them around a speaker, or turn an essay into a narrated film. With no footage, it generates new material. It studies a reference's pacing, typography and transitions before it plans, and checks the cut before you see it, according to the release post.
Where it lives
Motion connects to Google Drive and Frame.io for footage, and pulls context from Notion and Slack. It can run on a schedule or when a new YouTube video lands, then publish through connected social accounts. Work can be finished in the timeline, exported to Premiere Pro, After Effects or DaVinci Resolve, or driven from Claude, Codex or another agent over an API and an MCP server.
The framework argument
Mosaic's post makes a case rather than announcing a single feature. Tools like Remotion (React), Hyperframes (HTML) and plain Python all promise to make video with code, and their builders argue the language shapes the result. Mosaic argues the model plans the film in latent space first, so the framework matters less than the idea, and offers Motion as one more way to write that plan out.
Motion first launched on June 18, 2026, as a motion-design agent; the October 20 post is Mosaic's framing of it as a general video agent. Mosaic, a Y Combinator W25 company, had raised a $3.8 million seed round to build video editing agents.
For creators, the shift is from a generator you prompt clip by clip to an agent you brief once and direct in review. Whether the general framing holds up in practice is what hands-on tests will show; Mosaic has not published a benchmark for Motion's editing or video understanding.
Motion is available at motion.mosaic.so; Mosaic sells it on credit plans starting at $6 for 200 credits.
Sources
- mosaic.so - Mosaic's release post: capabilities, integrations, exports, the framework argument
- motion.mosaic.so - the product page and MCP/API access
- mosaic.so - the $3.8M seed round for video editing agents