Building an Agentic Video Editor for Mass Consumer — Ekaterina Deyneka, Reelful

Reelful’s agent analyzes uploaded media, transcribes speech, proposes a creative plan, and waits for approval. It then opens a **remote sandbox**, applies editing skills, builds a **Remotion composition**, and runs verification before rendering.

Builders can reuse this architecture for non-code artifacts: expose craft as skills, represent the output in an agent-editable format, and add a deterministic validation loop. A conventional editor remains available for precise human corrections.

Selecting the best moments from messy footage is more constrained than generation from a blank canvas. The product is still early, and the talk provides examples rather than measured edit quality or verification reliability.