"Why isn't there a Cursor for video editing?"

I’m putting together some AI video editing workflows, and in that process I keep running into the same three problems: tool calling, transcription and taste.

After long shoots, I might have 40GB of 4K S-Log footage, all in files titled something like “C0141.mp4”. During shoot, I always put together a rough narrative. It’s like I’m already editing the clips in my head, but it takes a while to turn that vision into a timeline.

Alongside my work with tech startups and my own business, I also work with political figures and other non-profit organisations. As you might expect, a lot of that work involves clipping good moments from speeches, interviews and other footage.

When I was working exclusively with them last year, I did much of this manually. The software engineering side of me kept thinking: “These are often the same sequences with very similar keywords, and I’m doing the exact same process each time, so why can’t I automate it?”

With accurate transcripts, object recognition and tracking, AI can be quite powerful at finding strong moments in raw footage, podcasts, interviews and speeches.

The problem is that the tools available today are either too limited, too unreliable or too awkward to fit into a real editing workflow.

• Descript and the Underlord MCP are probably my favourite setup for this purpose right now. They’re good for getting the basics out, but Descript is expensive, limited creatively and unreliable when it comes to tool calling. It can produce a starting point, but rarely a standout video.

• Palmier Pro looks incredibly promising, and I’m excited about it. But whenever I’ve used it, I’ve experienced long hangs. Its local transcription can also be off by a few seconds, and outside the AI-specific features, its colour-grading tools feel limited.

• Some people I know use Remotion. With the right /skills, it can be amazing. I sometimes use it for motion graphics, especially when a video has already been planned and coordinated. But Remotion is closer to a code-based version of After Effects than a traditional video editor. If you’ve edited video manually, you probably wouldn’t assemble an entire edit in After Effects. It feels similarly unnatural in Remotion.

• I’ve also tried Adobe Premiere Pro MCP plugins. They’re among the best options if you’re willing to watch the computer constantly, but they can’t access Premiere’s native transcripts. Instead, they run a separate transcription process, which slows everything down.

• If Adobe weren’t such a vast, bureaucratic corporation, I suspect we’d see more investment in tools that actually take advantage of its very good transcription system.

Underneath all of this is a deeper problem: AI models generally lack taste. This isn’t limited to video. Models are trained to treat everything like a PowerPoint presentation. They over-explain, over-structure and lack character. That lack of taste carries over into the emotional judgement behind which moments get clipped.

When it comes to building AI software for video editors, I think there should be two kinds of systems:

Fully automated, tightly scoped systems that clip existing content and adapt it to different audiences and formats. There is huge potential here for anyone creating mass-produced content, like newsrooms. Could the BBC have an agent monitoring its live news feed or media library 24/7, turning relevant moments into social clips? Most of the software I’ve mentioned above can’t cater to this kind of workflow, with the exception of Descript.

A real co-editor that organises media for me and places it in a timeline. I want to be able to insert footage, have it produce a high-quality first draft, then open that draft in the tools I already use so I can refine it and strengthen it. Think of an AI coding extension in VS Code, but for Adobe Premiere. This is the problem existing tools are currently trying to solve.

I believe this approach is what many high-end video agencies want: another interface that allows them to do their jobs faster.

Some companies will try to replace the video editor entirely, but I still don’t think AI can be creative to the same extent as a human.

People I know who use it to create videos still have to prompt it constantly, and much of this job is about judgement. As these tools become more powerful, the work shifts towards instinct, judgement and taste, rather than technical execution.

If “vibe editing” follows “vibe coding”, we’ll see many more people producing videos without learning the underlying craft. Better tools will make editing more accessible, but they won’t automatically give people taste, instinct or structure.

I’ve heard people in tech wonder why video editing hasn’t already seen the same kind of disruption as coding. The obstacle is that industry-standard tools like Premiere and DaVinci Resolve don’t offer the integrations or APIs needed to use the full capabilities of their software in agentic workflows.

Tools like Palmier Pro are trying to fill that gap by building their own custom tooling, but right now they still can’t deliver the level of control that someone using professional editing software expects.

Until then, there is no Cursor for video editing. There is only a growing collection of clever tools that still require too much supervision, or too limited, to become part of a professional workflow.