3 a.m., completely sleepless: Seedance 2.0 shows us that AI's “compression” of real-world workflows is accelerating
```
Author of this article: Shen Siqi
Source: Hard AI
At three o'clock in the morning, I saw Tim from Film Hurricane update his video on Byte Dance's Seedance 2.0, and I couldn't sleep at all.
For the first time in over a year, the advancement of AI made me feel this excited. Or, trembling.
Many people are waiting for the GPT-3.5 moment in the video field, and everyone thinks it'll take two or three more years. Seedance 2.0 tells us it's already right in front of us.
Its strength lies in turning camera movement, shot segmentation, and audio-visual matching all into AI—and it does them beautifully. It understands light and shadow, perspective, cinematic language.
What Tim shows in his video is control, a perfect AI replication of the physical world.
The logic of AI is becoming clear and simple, AI is frantically compressing our workflows: from directing, filming to editing, and scoring; from product managers, development to testing, and delivery.
All the intermediate steps are being gradually compressed.
In this article, I want to discuss how AI is changing workflows and reconstructing our work.
01 Video Industry'sGPT-3.5 Moment
I resonate with Tim’s uncontrollable excitement in the video.
Previously, we thought camera movement belonged to the physical world—sliders, cranes, drones, Steadicams. These devices are expensive, and the operators are even more expensive.
Seedance 2.0 turns all these into parameters. In the video demonstration, image-to-video: a photo of the protagonist + a photo of the scene.
It allows the protagonist to move within the scene according to your specified camera movement, and multi-subject consistency is astonishingly maintained.
Push, pull, pan, tilt—all previously required tracks, lighting technicians adjusting each second's lighting position.
Now it's just a line in the Prompt; the physical constraints of the physical world are replaced by the constraints of mathematical parameters.
Seedance 2.0 seems to understand the consistency of three-dimensional space.
It knows, for example, when the camera moves left, how objects in the background should create parallax. It knows when light shines from the right, how the length of the shadow should change.
Seedance 2.0 is beginning to touch editing: AI can understand video rhythm, identify emotional peaks in the frame, automatically match the beats of the music.
For editors, what used to take hours of “rough cutting” may now take only a few seconds.
Sound is the same—the image shows a basketball court, complex sounds from the arena appear synchronously.
This consistency of perception is a crucial basis for the human brain to judge “reality”; AI has achieved it.
Film post-production used to be an extremely complex systems engineering challenge. Director conceives, cinematographer turns concept into light and shadow, editor recombines visuals into narrative, composer manipulates emotions through sound.
This is an extremely expensive, inefficient, and friction-filled linear workflow. Seedance 2.0 shatters this chain and compresses all these jobs into one model.
Essentially, what AI is doing now is continually compressing various workflows.
From Seedance 2.0 we see the prototype of AI compressing the workflows of directors, filming, editing, scoring.
The GPT-3.5 moment in the video field has arrived.
The next two or three years will be the industry's reshuffling; the old order is collapsing.
02 AIIs Radically Compressing Our Workflows
The revolution in video is just one aspect of AI reshaping workflows; deeper change is happening in software, happening on our smartphone screens.
Recently, I used Alibaba's Qianwen to order milk tea, and this experience made me think a lot.
It may signal the end of the App era, or the arrival of the “instant software” era.
Our current internet experience is locked into the “App” form.
If you want to buy milk tea, you have to unlock your phone, find the food delivery app, click to enter, wait for the splash ad, click the search bar, enter “milk tea”, browse a list of dozens of stores, click into a store, pick from dozens of items, choose sweetness, ice, place order, pay.
This is an extremely long chain.
Why do we go through this process? Because Apps try to meet everyone’s needs, find the greatest common divisor, have to put low-frequency needs into secondary pages, have to add all kinds of recommendations for commercialization.
For me, I don’t need these—I always order from the same three shops, I know which lemon tea tastes best, which kitchen is cleanest.
I just need: “Order me a cup from my usual place, no sugar.”
Qianwen now is approaching this ideal state.
You give it an instruction, it calls interfaces via code and Agent in the background, and completes the delivery.
This is the “intent interface”—you express an intention, AI delivers the result. The UI, interaction, navigation in between are all compressed.
When AI's capability evolves from Andrej Karpathy's “Vibe Coding” to sufficiently powerful Agents, every need will be fulfilled via an instantly generated “one-time App”.
The traditional “product manager requirements document – developer writing code – tester finding bugs – final delivery” chain that can last weeks or even months will be compressed instantly to less than a minute by AI.
This leads to a fundamental business question: If I can generate an “App” in one minute to meet my current need, why download an app that's hundreds of megabytes?
The current App ecosystem suffers an insoluble structural contradiction: each person’s needs are unique, and AI can directly turn users’ natural language requirements into delivery via instant code.
Essentially, the AI tailor-makes a “personalized App” for the user—use it and leave, no need to retain.
This poses a huge challenge to current internet giants. Their moats are built on app installation numbers and user time spent.
If Apps disappear, if the entry point becomes AI Agent, where will their traffic come from? Where can they place ads?
The entry point for the next era seems to be becoming clearer.
Why all the big companies are crazily making large models, fighting for the answer to the only “super Agent” is evident.
Will many App products based on aggregated demand shift to niche AI-internal products in the AI age?
Current app developers may turn into “data API service providers”, and as the delivery chain is greatly compressed and costs drop, App demand essentially becomes API demand.
Every product conversation is a personal delivery as a product manager.
Ultimately, the disappearance of traditional workflows means the disintegration of company organizations.
The company as an organization exists fundamentally to reduce transaction costs. Communication is expensive, trust is expensive. So we gather people together, sign contracts, pay wages.
When one person + AI can accomplish what a whole team used to do, giant organizations become unnecessary, and we will see more and more “one-person companies”...
Seen from this perspective,
I believe, AI’s change to the world is accelerating.
This article is from WeChat public account “Hard AI”. For more AI frontier news, please go here
Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it take into account the special investment goals, financial circumstances, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable to their particular situation. Invest accordingly, responsibility is your own. ```