Over the past two days, Seedance 2.0 has gone completely viral.
In the view of Feng Ji, founder of Game Science, Seedance 2.0 brought about important changes. All presentation methods that previously required multiple production cost considerations will soon be “video-enabled.” E-commerce advertising, branded materials, and pre-shot content will be affected first.
So how will AI reshape video workflows when production barriers are removed?
Today we’re highlighting key insights in the AI video space from a16z partner Justine Moore. As one of the most active early-stage investors in the application layer of AI in Silicon Valley, she has led investments in many groundbreaking projects such as ElevenLabs and Krea, consistently publishes consumer-grade AI trends reports annually, and demonstrates strong forward-looking judgment in the evolution of creative tools.
Justine’s core conclusion is that the real change in the next stage is not the generative layer, but the “editing layer.” And AI agents are quietly evolving into invisible “post-production teams.”
In her view, three conditions matured around the same time. One is the ability of large-scale visual models to understand the semantics and narrative structure of content. The second is the scheduling and collaboration capabilities of multimodal tools. Third, the stability and aesthetic quality of the generative model has been dramatically improved.
When these three points cross a critical threshold at the same time, AI is no longer just “providing the material” but begins to adjust the process, refine the details, adjust the rhythm, and even shape the taste to some extent. A workflow centered on “AI editing agents” is taking shape.
Next, we analyze this technological inflection point from five aspects. How exactly will AI agents reconstruct the complete chain of video creation and why it will become the next real competitive high ground.
01
When the AI video explosion encounters a creative dilemma
2025 is known as the “Year of Video.” AI-generated ads are going mainstream, and some seed-stage startups’ launch videos can garner millions of views. Video podcasts and interviews are also experiencing explosive growth, with dynamic images occupying ubiquitous screens.
But behind this prosperity lies a long and tedious behind-the-scenes job. Refining 90 minutes of raw footage into a 3-minute short film. Lighting and sound are painstakingly modified in post-production. Continuing to search for the perfect sound effect is part of the daily routine of video production.
There’s an “80/20 rule” for video production: spend 80% of your time and energy on editing and 20% on filming (this generation). This is a test of your sense of how to tell a story, how to control the rhythm, and how to move the audience. Creating truly engaging videos is still a painstaking process that requires a great deal of patience and professional judgment.
There is now technology to delegate some tasks to AI agents, which can help with the production of filmed and generated content. Large-scale visual models can be viewed and understood through large amounts of video footage. Agents can analyze, plan, and use editing tools on your behalf. There is enough training data to teach the model how to make good videos.
AI video agents will significantly increase the supply of high-quality videos. This type of content currently takes days or even weeks to produce by professional video editors. Just as Cursor revolutionized programming, these agents will revolutionize video production.
02
How can AI take on the “troublesome and troublesome work” of video editing?
There is a huge market demand for AI agents that allow anyone to acquire the skills and flair of a professional video editor. So why aren’t such products popularized yet? Several recent developments are driving change.
Large visual models can now handle large amounts of video. You need to understand your video before editing. This is no easy task. Even a very short clip requires processing a large amount of information.
Significant advances have been made in recent large language models such as Gemini 3, GPT – 5.2, Molmo 2, and Vidi2, which are inherently multimodal with long context windows.
Gemini 3 can now handle up to an hour of video. You can upload this as input and have the model generate timestamp labels, find specific moments, or simply summarize what happened.
The models learned how to use tools. AI editors need to be able to perform actions, not just make suggestions. We have seen significant progress in large-scale models where the tools act as agents that can be used in practice.
One of my favorite examples is Claude using Blender (3D creation software), a complex tool that many people find difficult to master. You can imagine the possibilities that would open up if agents had more tools at their disposal.
Improved quality of image and video generation models. I am convinced that the future video production process will be hybrid – Combine AI-generated content with real-world content.
Imagine filming a documentary interview and using AI to generate establishing shots and historical images. Or use motion transfer models to apply animation references to real-world characters. For these methods to be truly useful, the model must meet certain quality and consistency criteria. And now this is becoming a reality.
What can these AI agents do?
Below are some examples of the types of tasks they can handle for us.
First, process management. The amount of final footage, whether shot or generated, is often far greater than needed (sometimes hundreds of times more; think of how many “alternate takes” a movie or TV series has).
Organizing, reviewing, and deciding which footage to use is often difficult. Products like Eddie AI can process hours of uploaded video, identifying the main shot and establishing the shot, processing multi-angle camera positions, and comparing shots.
The second is multi-model orchestration.. If many videos in the future will include AI-generated elements, you’ll need an agent that can coordinate all the models.
For example, adding AI animation to an educational video requires an agent to generate images, send them to a video model, and stitch the output together. Products like Glif launch agents that can coordinate work between multiple models on your behalf.
Third, improvement of details. To take your video from “so-so” to “excellent” you need to fix small details.
But if you’re not a professional editor, the sheer number of tweaking tasks can be overwhelming. For example, you can adjust the lighting between clips, remove noise from audio tracks, or remove filler words like “uh” and “ah” from interviews. Products like Descript’s Underlord agent can take over the video, make all these changes, and deliver the final version.
Fourth, format adjustment. After video production is complete, adjustments are often needed to expand your reach.
For example, edit your YouTube podcast into short videos in different aspect ratios and post them to your X, Instagram, and TikTok accounts. Or even translate and redub your videos to reach a global audience. Platforms like Overlap allow you to set up node-based workflows for these adaptation tasks.
Fifth, taste optimization. The ultimate goal is not only to replace manual work with AI, but also to train smart agents who can improve the quality of videos.
There are reasons to hire a professional editor. Because it makes the photo more beautiful. They spend years learning how to engage an audience, control rhythm, and evoke emotion with their music. This involves thousands of small decisions.
YouTuber Emma Chamberlain once said that it took her 30 to 40 hours to edit a 15-minute video blog.
Imagine if an AI agent could watch your video, ask you about your goals, and then generate a few draft edits that you could iterate on. All you have to do is provide feedback such as “starting too slow,” “cut the middle,” or “make the ending more impactful,” and the agent will take action.
Video has become mainstream. It’s how we learn, market, and connect. However, editing bottlenecks are becoming increasingly evident. We need to capture more footage, post it to more platforms, and adapt more formats.
The good news is that technology is in place to solve the problem. Visual models, agent-based tools, and large amounts of training data have all matured over the past year. All the pieces of the puzzle are ready.
This means that AI editing agents will significantly improve the quality of every video we watch and significantly speed up production in the coming months and even years.
This article is from the WeChat official account “Silicon Base Observation Pro”, Author: Silicon Base June. Republished by 36Kr with permission.
