An AI agent that edits the video for you

Editing a 30-second video still takes 30 minutes

Traditional video editors expect creators to do everything manually: choose a style, set transitions, sync to audio, fix levels — then maybe export. The result is friction, abandoned drafts, and inconsistent quality across reels and shorts.

Even AI-assisted tools today push the user back into a timeline. They help with one step. They don't take responsibility for the outcome.

Mapping the messages users actually send each other

We collected real conversations from creators planning a video together. Friction lives in iteration loops: 'shorter', 'remove the boring part', 'add a hook at the start', 'try a different vibe'. Those are agent commands — not editor controls.

From this we extracted a small set of recurring intents: trim, restyle, hook, music, captions, ratio. Everything else is edge-case.

An agent that takes initiative — you only approve

Instead of a timeline, the user gets a conversation with an editor. The agent reads the raw footage, proposes a first cut, explains what it did, and shows the result. The user replies in plain language; the agent re-edits.

Three loops: ingest → propose → refine

The agent runs on three nested loops. Ingest: scene detection, transcript, audio analysis. Propose: a first cut with a stated intent. Refine: respond to natural-language requests with new edits and a clear changelog.

Two surfaces: chat and preview

The mobile UI collapses to two surfaces only: a video preview that always shows the latest cut, and a conversation thread where the user nudges the agent. Every agent action is reversible by replying.

For longer projects we add a third surface — a 'shots board' the agent maintains automatically. The user never builds it themselves; they only mark a shot as 'must keep' or 'cut'.

A direction, not a destination

This case is an exploration of how agentic UX changes the shape of a creative tool. Not every product needs an agent — but when iteration speed is the bottleneck, removing the timeline can be more valuable than improving it.