ChatCut is an AI video editing product. Its founder, Kaiwen Li, is a former VICE and Discovery commercial-documentary director whose short film Seed Hunter was shortlisted for the Golden Horse Awards. It originally targeted the most painful cut for documentary/reality-show/podcast creators — transcribing interviews, finding the best lines, reordering, and assembling a rough cut of the story. After eighteen months of evolution, its positioning made a key migration: from "a standalone editing app" to "the professional video execution layer behind general agents." In July 2026, ChatCut launched Codex and ChatGPT plugins, which means you can describe your intent in natural language inside a general agent and let ChatCut turn it into an inspectable, modifiable video project.

What It Is: An Eighteen-Month Evolution Path

To understand ChatCut's value today, you first have to see how it got here step by step. The unit of competition (the smallest granularity users pay for) kept changing:

The Starting Point: Winning the Most Cognitively Expensive Interview Rough Cut (PaperCut)

Documentaries, reality shows, and podcasts pile up interview footage, and directors have to transcribe it, find the best lines, reorder, and assemble the story. ChatCut 1.0's interface put Bin Viewer on the left and PaperCut on the right: select, mark, strike through, and reorder sentences in the source transcript while the video follows the text, then export XML for Premiere and DaVinci Resolve back into professional software for finishing. It makes only one cut — but the most cognitively expensive, most repetitive one.

Evolution: Language Becomes the Control Plane

From September 2025, "editing video with text" became "making editing feel like chatting." Import three long videos at once, describe the story structure directly, and the AI returns a plan and writes the segments onto the timeline. The chat box did not replace the editor — the transcript, timeline, drag-and-drop, subtitles, audio, and XML are all still there. Conversation expresses intent, the editor exposes state, and the human verifies and refines.

Maturity: The Complete Task Becomes the Unit (General Editing Agent)

In a February 2026 screen recording, it completes an entire chain: after a YouTube fetch fails it automatically switches to a media downloader, then extracts frames, reads the transcript, scrapes the web page, analyzes style, plans a storyboard, generates assets, and finally writes back to the timeline. The software no longer waits for users to press buttons — the agent breaks the goal into steps, picks tools, handles failures, and delivers a project you can keep editing.

How to Use It: Cutting a Video Inside Codex / ChatGPT

After the Codex and ChatGPT plugins launched in July 2026, the entry point moved outward — the user first states the goal inside a general agent, Codex presents the content and editing judgment, and the ChatCut workflow writes cut points, MG (motion graphics), music, and sound effects onto the timeline.

Hands-On Workflow

One creator ran a round inside Codex, and nearly all of the work happened through conversation:

1. Enter the goal in Codex: describe the video content, style, and pacing you want.
2. Codex presents the content and editing judgment: it analyzes the material and proposes an edit plan.
3. Write to the timeline via the ChatCut workflow: cut points, MG, music, and sound effects land on the timeline automatically.
4. Refine in the professional interface: the timeline, transcript, and subtitles are all there; a human verifies and fixes the flaws.

Hands-On Numbers

• Efficiency: editing/subtitle/sound-effect/BGM work that used to take 40 minutes in CapCut was done with a few chats in Codex.
• Cost: effectively free (it never touched ChatCut's own generation capabilities — it went through the Codex + ChatCut workflow path).
• Flaws: edit breathing pauses have minor issues; some transcribed text needs a manual pass (for example, "Agent" was recognized as "A juan" — a Chinese homophone).

💡 Tip: ChatCut's core trade-off is that "getting to 80% but leaving room for humans to fix it" beats "getting to 90% but not letting anyone change it." After generating a video inside a general agent, always go back to the timeline and manually review the breathing pauses and the transcription — that is the universal weak spot of AI video editing today.

Why the Entry Point Moved Outward While Professional Capability Became More Valuable

A counterintuitive judgment: the entry point being siphoned off by general agents does not mean professional capability loses its value. Quite the opposite — the more complex the tasks agents take on, the more the layer behind them needs to be reliable, transparent, editable, and hand-off-able.

Kaiwen Li draws a new moat standard for professional software:

• Callable: general agents can invoke it through plugins/workflows.
• Observable: every operation is visible to the user, not a black box.
• Editable: what it produces are project files (XML, timeline), not dead video.
• Takeover-able: a human can step in and modify at any time, never locked out by the AI.

The cost certainly exists: if users only remember "I edited a video in ChatGPT," the user relationship, brand mindshare, and pricing power may be captured by the general agent. What ChatCut has to prove next is whether it can be more reliable, more controllable, and more engineering-literate than other execution layers.

Use Cases

• Documentary/reality-show/podcast rough cuts: ChatCut 1.0's founding scenario — transcribe, find the best lines, reorder, export XML back to Premiere/DaVinci for finishing.
• YouTube long-video repurposing: fetch a long video, then automatically extract frames, read the transcript, plan the storyboard, and generate short-video assets.
• Batch output for content teams: describe intent in Codex/ChatGPT, land the timeline via the ChatCut workflow, and leave humans only the final polish.
• Ad films/motion graphics: the AI motion-graphics generation capability added in November 2025 can recompose footage into new visuals.

FAQ

• Do you have to install ChatCut to use it?: the Codex/ChatGPT plugin path does not require opening the ChatCut app on its own, but for finishing we recommend its professional interface or exporting XML to Premiere/DaVinci.
• Can it fully replace CapCut/Premiere?: no. Today the AI finishes 80% of the grunt work; breathing pauses, transcription, and finishing still need a human. Its position is the execution layer, not a replacement for a professional NLE (non-linear editor).
• Which export formats are supported?: XML project files for Premiere and DaVinci Resolve, making it easy to return to professional software for finishing.
• Is the transcription error rate high?: proper nouns (like "Agent" or brand names) are occasionally misrecognized; give the subtitles a manual pass before final output.