Feature Requests

Feature Request: Agent Underlord Access to Comments & Notes Problem: Currently, Agent Underlord cannot see comments, notes, or annotations that users add within Descript (DRIVE) projects. This creates a workflow gap where users must manually transcribe their editing notes into the chat instead of having the agent read them directly. Requested Feature: Enable Agent Underlord to read and reference comments and notes added to Descript projects, including: Timeline comments/annotations Script notes and inline comments Project-level notes or task lists Time-stamped feedback markers Why This Matters: Natural Workflow Integration Users already document their editing decisions using Descript's native commenting tools. Requiring manual duplication into chat breaks the flow. Efficiency Eliminates the need to re-type or screenshot notes, especially for complex projects with multiple change requests. Context Preservation Comments often include timestamp references, visual markers, and rich context that's difficult to communicate through text alone. Collaboration Alignment Many users work in teams where comments are the primary feedback mechanism. The agent should participate in that same communication layer. Example Use Case: A user reviews their 15-minute video, adds 8 timestamped comments like "cut this section" or "add title card here," then asks Agent Underlord: "Can you make the changes from my notes?" The agent would read all comments and execute the requested edits without requiring manual re-entry. Expected Behavior: When a user references "my notes" or "the comments I added," Agent Underlord should be able to query and retrieve comment content, timestamps, and context, then execute the requested changes accordingly.

I work on several podcasts, and my participants don't have great microphones so I lean pretty heavily on Studio Sound to clean up the audio. But Studio sound will somehow leave in tongue clicks, or lip smacks, (which I have to edit out) but then eliminate laughter. Why? I want to hear them laughing at each other's jokes, etc. Can we please train it to leave the laughs in? Thanks!

The tool is absolutely useless. The AI is constantly making mistakes and regularly destroying the entire timeline, cooking whole parts without being asked, changing times and speeds unintentionally... it is an ABSOLUTE disaster. AI level from 2 years ago... isn't worth the effort to upload your material here. Absolute waste of time and money.

since last Wednesday its hasn't been detecting any other filler words than um & uh, my transcript is in english, it's that same as before, I'm on the newest version of descript, they said the only way to get this handled was to put is in here so fingers crossed this works, please fix this AI function.

Whenever the Underlord gives me an answer, I want to start reading this answer right away, without waiting until the answer is completely finished. But currently I'm not able to do this, because when the Underlord appends text to the bottom, the entire paragraph of text moves up, which makes it not readable. Usually when this happens on any other web-based chat application, I'm able to scroll up a little bit so that expanding the page to the bottom doesn't change the position of the paragraph that is being appended to. But in Descript application, whenever I do this, it forces to scroll to the bottom and the paragraph just moves, making it unreadable.

After editing just one video all my monthly credits are gone. This is insane. Can you please either adjust credit price by x10 or revise how you account for their usage.

The Underload feauture should be able to keep a history of prompts that were inputed in the past, so that when users come back to the project, they remember where they left off

Automatic camera versus screen share switching on multi-track with Underlord based off of mouse and computer screen activity Basically, I want the Underlord to be able to tell when I'm moving my mouse, when I'm clicking stuff on screen, all that sort of stuff on my screen share. If I'm not moving my mouse and not clicking my screen, it'll go back to my full face camera, where it shows my full face, so the editors can add animations, stuff like that. If I'm doing stuff on my screen, it goes back to screen share.

It just doesn't work. Please stop trying to make fetch happen. Please build useful tools for video editing, not a terrible AI layer that can't do anything.

My eyes are jumping all over, particulary one eye. This file is in Coaching Lectures, inside is Mod 1 and Mod 2 for example.