Agentic editing is easier to see than to define. So here is one edit, step by step: a raw interview, an AI assistant connected to Arsaze, and a person who wants a short, captioned video at the end. The assistant could be Claude, ChatGPT, Grok or the built-in Arsaze assistant. The steps are the same.
The brief: "Turn this interview into a tight three-minute story. Keep the best answer about why we started. Add captions and clean up the sound."
1. It reads the project
Before it touches anything, the agent reads what is there: the clips on the timeline, how long they are, and the transcript of everything said, with who said what. This is the agent's version of watching the rushes. It is also why a real timeline matters: the agent sees the whole project, not just one file.
2. It looks where it needs to
A transcript does not say whether someone is in frame or looking away. When the agent needs to know what is on screen at a moment, it looks at that moment. In a local project, your files stay on your computer while it does, and nothing is kept.
3. It plans before it cuts
Ask the agent to show its plan first: which answers make the story, in what order, and what gets cut. You can read the plan and redirect it before a single edit is made. "Keep the story about the first customer" costs nothing at this stage.
4. It makes the cut
Now it edits. It removes silences and filler words, drops the retakes, keeps the best take of each answer and puts the answers in story order. Each step is an ordinary edit on the timeline, playing back live if you have the project open. If you would rather read than watch, the Rewrite view shows the transcript as a document, so you can see what is left in.
5. It adds what the story needs
Captions, word by word, in the style you asked for. A title card. Music under the voice, ducked so the speech stays clear, and the loudness evened out across the video. On a paid plan it can also generate b-roll, images or a voiceover and place them on the timeline. On any plan it can use clips that are already in your library.
6. It checks its own work
This is what makes it an agent rather than a one-shot tool. It compares the result with the brief: is it about three minutes, does the story still make sense where two answers meet, are the captions in time with the voice. When something is off, it fixes that part instead of starting over.
7. You review it
Watch the cut and change what you want, by hand or by asking. Every edit the agent made is in version history, so going back to an earlier state is one step. For a client's view, share a review link. They leave comments on exact frames, and the agent can read those notes and make the fixes. Review links need a cloud project.
8. It exports
When you are happy, export an MP4. You can also export the edit as an FCPXML project and keep working in DaVinci Resolve or Premiere Pro. Exporting on your own computer with the desktop app is free and has no watermark.
Where your time goes
Two moments matter most: the plan in step 3 and the review in step 7. A clear brief and a few minutes on the plan save more time than anything else. The agent does the slow, repetitive work in between, and you keep the decisions that make the video yours.
Try it on your own footage
Connect your assistant, Claude, ChatGPT or Grok, open a project and give it a brief like the one above. For the bigger picture, read what agentic video editing is.
Questions people ask
What does an AI agent do when it edits a video?
In Arsaze, an agent reads the project and its transcript, looks at the moments it needs, plans the cut, makes the edits on a real timeline, adds captions, music and sound fixes, checks the result against your brief and fixes what is off. You can watch every edit live and undo any of them.
Does the AI watch the whole video?
It reads the transcript of the whole video and looks at specific moments when it needs to know what is on screen. In an Arsaze local project your files stay on your computer while it looks, and nothing is kept.
Can I change what the AI edited?
Yes. Every edit is an ordinary edit on the timeline that you can change by hand, and every edit is saved to version history, so you can go back to any earlier state.
What should I put in a brief for an AI video edit?
A few sentences are enough: how long the video should be, what it is about, what must stay in, and the style of captions and sound you want. Specific briefs give better first cuts.
Can a client review a video an AI edited?
Yes. Share an Arsaze review link and the client can comment on exact frames. The AI assistant can read those comments and make the fixes. Review links need a cloud project.