Why AI Won’t Replace Your Editor – But It Will Change Everything

Andrew Ford is a software engineer in San Francisco who took a year off from his career to go all-in on YouTube. He’s also a trained journalist. He knows how to write a script, he got comfortable on camera, and he had a genuine story to tell. But editing nearly broke him — including one video he spent five to six months finishing. That experience led him to build Buttercut, an open-source AI editing tool built on top of Claude that turns raw footage into a structured rough cut without you having to scrub through hours of clips at midnight.

This conversation cuts through the usual AI hype to get at what’s actually changing in video editing, what’s not, and what that means for creators and editors trying to figure out where to put their energy.

What AI Editing Is Actually Good At (And What It Can’t Touch)

The honest answer to “can AI edit my videos?” is: sort of, and only if you set it up correctly. Buttercut works by giving Claude a structured framework for processing footage — first transcribing everything, then analyzing and summarizing each clip, then building a library that the AI can reference all at once. From there, you tell it what kind of rough cut you want.

The real value isn’t magic. It’s digest speed.

“AI can just read every single transcript and look at every single timeline, like all at once. So it can kind of just almost instantly understand it in a way that it’s a lot harder for us.”

Anyone who edits their own footage knows the specific pain of sitting down after two weeks away from a project and spending the first 45 minutes just remembering what you shot. AI doesn’t have that problem. It can hold the entire shoot in working memory simultaneously — something no human editor can do with a multi-hour raw dump.

The other thing AI does surprisingly well is breaking creators out of tunnel vision. Andrew described shooting a vlog with what he thought was one obvious linear structure. When he ran it through Buttercut, the tool offered multiple narrative options — including starting with the payoff and working backward. Not something he’d considered. Having an assistant that can propose alternate story structures without ego or attachment turns out to be genuinely useful, even when the output isn’t final-cut-ready.

Where AI Editing Still Falls Apart

Pacing. Full stop.

Andrew is direct about this: AI does not have an innate sense of what feels right rhythmically. It won’t notice that a clip is two frames too long. It won’t instinctively cut to B-roll at the moment a viewer’s attention is about to drift. These aren’t small problems — pacing is what separates watchable video from something that feels slightly wrong in a way you can’t articulate.

“There’s like a natural cadence to pacing that feels nice… that kind of naturally comes to humans, but that does not come naturally to AI or Claude right now.”

The workaround exists: you can write explicit pacing rules into Buttercut’s instructions. Tell it to hold on the speaker for three seconds, cut to B-roll for 1.5 seconds, return to A-roll, repeat. It will follow those rules reliably. But that means the rules have to come from you — someone with taste who already knows what good pacing looks like. The AI executes. It doesn’t originate.

And taste itself is the deeper limitation. Andrew tested Buttercut against Claude, Codex, and Gemini. Claude came out ahead on editorial judgment — but even the best model is nowhere near a human editor who genuinely cares about the work. The analogy he draws is useful: when a developer uses AI to write code and the feature works, you can’t tell the difference. But when AI assembles a video edit, anyone with decent taste can feel that something’s off. The texture isn’t there. The decisions are too average.

The Real Bottleneck Isn’t Editing — It’s the Story

One of the more interesting threads in this conversation is where Andrew thinks the bottleneck moves once AI handles the mechanical work of cutting footage. His answer: story and script quality become the constraint, not editing bandwidth.

He points to Hollywood as a preview. There’s no shortage of skilled actors. What’s scarce is a script worth performing. You can see this in real time — a talented cast in a mediocre film versus the same actors in something with a strong story. The script was the variable. Andrew thinks YouTube is heading the same direction as AI removes editing as the chokepoint.

This framing changes how creators should be investing their time right now. If you’re spending 60% of your creative energy on editing logistics, and that gets automated, you don’t get to coast. That freed-up capacity needs to go into developing better story instincts, stronger premises, and cleaner scripts. The tools lower the floor. They don’t raise the ceiling for you.

Andrew’s Viral Video — What Actually Made It Work

Andrew’s most successful video — currently sitting at around 600,000 views on a bike channel — wasn’t an accident. The premise was tight: he bought an expensive e-bike in San Francisco (a city with serious bike theft problems), it didn’t come with GPS, and the video explored whether a stolen bike could be recovered if it had a tracker on it. High-stakes, clear hook, relatable frustration for the target audience.

But it didn’t blow up immediately. The first three months, it topped out somewhere between 500 and 1,000 views. What changed? He tested 15 to 20 different thumbnails before the video started getting traction. He’s honest that he doesn’t know for certain whether the thumbnail was the actual unlock or if it would have eventually found its audience regardless — but the testing mattered. Thumbnails and titles are how you communicate the premise before someone clicks. If that preview doesn’t land, the video doesn’t get a chance.

His advice on titles is simple: shorter is better. When you’re scanning a YouTube feed, you’re not reading — you’re pattern-matching. A clear, short, active title that communicates the premise immediately outperforms a clever long one nearly every time.

AI Slop Is Real — And Audiences Are Already Filtering It Out

Andrew watches a lot of camera gear content on YouTube. He’s noticed a surge of videos that are clearly AI script, AI voiceover, AI B-roll assembled into something that looks like a review. His response: he blocks the channel entirely. Not hides the video — blocks the channel. He has an hour or two for YouTube each day and has no interest in spending it on content that a person didn’t meaningfully make.

What’s interesting is that he also noticed plenty of comments on these AI-generated videos from viewers who seemed not to notice. So the detection rate isn’t universal — but it’s growing. And the viewers most likely to catch it are the ones most worth having: engaged, discerning people who care about the subject matter enough to follow creators for years.

His diagnosis of why fully AI-generated content feels hollow goes deeper than prompting technique. It’s a structural problem:

AI has read everything. Every novel, every Reddit thread, every how-to blog post. Which means when it writes a script, it produces something that sounds like the aggregate of all of that — competent, smooth, and profoundly average. Or it skews toward marketing copy, because that’s heavily represented in training data. Human creators are more interesting precisely because they’ve only lived one life, seen one slice of the world, and formed opinions from that limited but specific experience. That limitation is the feature, not the bug.

What This Means for Video Editors Specifically

Andrew pushes back on the assumption that AI is going to hollow out editing as a profession the way it seems to be hollowing out software engineering. His reasoning: in programming, the output is functional or it isn’t. A button either works or it doesn’t. AI code that passes tests is indistinguishable from human code. But video isn’t like that — there’s no test to pass. There’s only whether it feels right to a human watching it.

His actual prediction is that AI could be a net positive for the editing industry, not a threat. There are creators out there right now — in niches like tiny homes, cabin builds, crocheting, van life — who have genuinely interesting stories but no time to edit. If AI reduces the editing burden enough that those people can publish consistently, the total demand for editing (and editing talent) increases. More creators in the market means more work, not less.

For agencies and freelancers specifically, his practical advice is to let AI handle the mechanical A-roll assembly when a script exists, and use it aggressively for structural brainstorming. What you protect is taste — the part that requires actually caring about whether the video is good. That’s where professional editors earn their rate, and it’s not something a model is going to replicate anytime soon.

What Creators Should Actually Focus on Building

The conversation ends on a question worth sitting with: if AI eventually automates execution, where does human value live?

Andrew’s answer is straightforward. Video is still a three-step process: script, shoot, edit. If editing gets handled, two steps remain — and neither of them has been solved. Nobody has ever finished a YouTube vlog and thought the script was perfect and the shooting was flawless. Those two disciplines have essentially unlimited room for improvement. Most creators also have a Notion document full of video ideas they’ll never get to. Clearing the editing bottleneck doesn’t create a content crisis — it creates room to actually work on the ideas that have been sitting there for months.

He also pushes back on the idea that AI tools mean creators need to dramatically increase their output. One strong video per week that earns genuine attention is still worth more than five mediocre ones optimized for volume. The tools should make your existing cadence more sustainable and your existing work better — not turn you into a content factory. His example: if Casey Neistat had published one good video a week instead of burning out on daily vlogs, he’d still have a massive audience. The frequency wasn’t the value. The quality was.

The answer to avoiding creative laziness with AI is the same answer it’s always been: have taste, expose yourself to work you genuinely admire, and have actual goals for where you’re trying to go. AI won’t make you lazy if you care about the work. And if you don’t care about the work, the laziness was already there — the tool just makes it more visible.

About Buttercut

Buttercut is free and open source as of this recording. Andrew has been building it since December and updates it frequently — he describes it getting meaningfully better every few days. The pricing will likely shift to a one-time purchase (he’s thinking around $30 to start, eventually up to $100), but right now it’s free to download and use with Claude. Find it at buttercut.io and join the Discord if you want to follow development.