Cutroom

Problem

Editing long-form footage is mostly search. An editor knows the moment they want and spends the session scrubbing to find it. Every tool sold as an AI editor so far has been a wrapper — it generates a rough cut you then have to fix in a real NLE, which means doing the work twice.

Approach

1. 01

   Transcribe on ingest, then treat the transcript as the primary timeline surface. Deleting a sentence performs the cut.

2. 02

   Build the actual editor underneath it — media pool, multi-track timeline, color, delivery — so the transcript view is an interface onto a real project, not a preprocessor that hands off.

3. 03

   Keep the model in an advisory seat. It proposes selects and finds moments; it never commits a change the editor did not ask for.

What it taught me

The hard part was never the model. It was making a timeline that stays responsive at hour-long durations while a second view of the same project stays in sync with it.