Descript Is Hiring an AI Researcher to Teach Video Editors How to See

If you're an AI researcher who cares less about publishing papers and more about shipping things real people use, Descript's open role is worth a look. The San Francisco-based company — which also hires remotely across the US — is building specialized models for specific editing problems, not broad general-purpose AI systems.

The work is unusually concrete for a research position. Descript's team has already put features like video translation, lipsync, voice cloning, and audio cleanup into the hands of millions of creators. The cycle from prototype to production has been measured in months, not years.

The new hire will focus on multimodal understanding — essentially training models to perceive video the way an experienced editor would. Right now, Descript's AI editing agent, Underlord, works mostly from text-based representations of a project. The goal is to give it genuine perception of the actual media: enough to judge whether a cut works, whether the pacing feels right, and whether creative choices land.

That last part is genuinely unsolved territory. There's no established method for representing or evaluating editorial craft algorithmically, which makes this a real research problem rather than an engineering polish job. Descript says it has a unique dataset to work with, which is often the limiting factor in applied media AI.

Inspired by reporting from We Work Remotely: Remote jobs in design, programming, marketing and more.