From Information Containers to Work Systems: How AI Products Shift from Information to Productivity
When designing AI products, the first challenge product managers face is often not a specific use case, model capability, or interaction pattern. It is something more fundamental - the object of design itself has changed.
In the internet era, products were designed around information. The core problem to solve was how information should be produced, organized, distributed, and consumed. As a result, product forms gradually converged into different kinds of information containers.
In the AI era, products began to directly carry productive capacity. The question is no longer how information should be presented, but how AI’s productive capability can be organized, invoked, and sustained over time.
Once the design object changes, the assumptions behind existing product methodologies and structural models begin to break down.
If we were to describe this shift with a simple analogy:
Internet products are like newspapers, while AI products are more like offices.
It reflects a fundamental shift in the design object, product structure, and value loop.
Internet Products Are Designed Around Information
The internet solved an information problem: how information is produced, organized, distributed, and consumed.
As a result, the design object of internet products was clear from the beginning - information itself.
The core responsibility of a product manager was to design information containers that fit specific contexts:
where information lives, how it is structured and distributed, and how users continuously consume it
Over time, information containers evolved through several clear stages:
In other words, designing internet products has always meant designing a newspaper. Even though the form of the newspaper changes, the design object remains information, and the design paradigm consistently revolves around information containers.
AI Products Are Designed Around Productivity
The emergence of AI is not merely about generating content faster. It introduces a callable productive force directly into products - one that can participate in task decomposition, path selection, execution, and result verification.
Under this premise, product managers face a fundamentally new question:
How do you design a work container that can host, schedule, and constrain this productive force?
This is the most essential difference between AI products and internet products.
Similarly, work containers have also gone through stages of evolution:
The real dividing line is not whether a product "has AI", but whether its container is designed for AI-driven productivity.
To answer what kind of container can truly support AI productivity, we must understand how humans work, how AI works, and how the two can collaborate within a shared structure.
The File System as a Shared Work Container for Humans and AI
Why File Systems Fit Human Work
Human work is not about producing one-off outputs. It is a continuous process of moving something from a historical state toward a target state.
Every step forward happens under constraints: progress toward a goal always comes with real cost.
Why File Systems Fit Human Work
The Temporal and Spatial Structure of State
Any working state exists simultaneously across two dimensions.
In time, it inherits from the past, exists in the present, and points toward the next step.
In space, it acts on concrete objects, with clear scope, granularity, and cost.
For work to progress continuously, states must be stably expressed, accessed, and operated on.
Files as the Minimal Expression of State
Files do not merely store content. They express state.
- Historical documents express completed states
- Active working files express ongoing states
- Strategy or goal documents express intended future states
Files make states visible, inheritable, and operable.
Folders as Containers for Managing and Advancing State
Folders are not just for organization. Their primary role is to manage the full context of a piece of work.
Within a folder, historical, current, and target files coexist, collectively defining scope, origin, and next steps. They stop being isolated content and become a continuous work state.
This does not mean file systems are the only way to advance work. But through long-term practice, they have become one of the most stable and widely adopted structures for organizing and advancing work since the birth of computing.
Why File Systems Also Fit AI Work
Once we understand the structure of human work, AI’s working logic reveals a similar - but more constrained - pattern.
How AI Works: Tokens and Context
At a fundamental level, whether generating text, writing code, or planning tasks, models always do the same thing:
given a context, predict the next token based on existing tokens.
An "output" is essentially a sequence of predicted tokens.
Whether the output meets expectations depends not only on how capable the model is, but on which tokens constrain it before generation.
Those context tokens determine three critical factors:
whether the goal is clear, whether granularity is controlled, and whether scope is well defined.
The Structural Constraint of Context: One-Time Windows
Context itself has a fundamental limitation. It is not a persistent workspace, but a one-time computation window.
This means that before every inference, the system must reconstruct an appropriate context for the model.
The Economic Constraint of Context: Token Cost
Context is also a cost-bearing resource. Every token participates directly in inference.
More tokens mean higher computation cost and latency. As a result, AI product design is not about giving models more information, but about constructing the smallest sufficient context within a limited token budget.
File Systems as an External State Space for Context
When work states are stably stored in an external system, context no longer needs to be fully loaded at once.
The system can selectively retrieve, trim, and combine relevant state to construct just-enough context for the current task.
The file system functions as this external state space.
Files and folders are not piles of information, but accumulated state representations centered around concrete work. They define clear object boundaries, establish explicit scopes, and allow historical and current states to be read together.
A Proven Structure: Coding Products
This structural advantage has already been validated in coding products.
Software evolves through continuous maintenance and modification of concrete code files. Each change is written back into the file system, and subsequent work proceeds from those states.
AI demonstrates sustained, controllable productivity in programming not because it is inherently "smarter" in this domain, but because code already exists within a highly structured, evolvable file system.
How File Systems Amplify AI Productivity
Looking back at AI’s working nature, file systems do not amplify intelligence. They amplify the probability that AI outputs meet expectations, and the likelihood that work can progress continuously.
For this reason, this design will not be "eaten" by stronger models.
Models get stronger. File systems ensure that strength lands continuously, economically, and reliably in the right place.
When Humans and AI Collaborate in the Same File System
When a file system satisfies both human needs for expressing work state and AI’s structural and cost constraints for context construction, collaboration fundamentally changes.
From Instruction Loops to State Handoffs
Collaboration no longer primarily happens at the conversation layer. It revolves around work state itself.
Files become shared working objects. Folders define shared boundaries.
Humans adjust direction by modifying goal and constraint files. AI advances execution based on existing state.
Collaboration shifts from instruction ping-pong to state-based handoff:
humans judge and validate, AI executes and advances.
From One-Off Outputs to Evolvable Work Assets
Once AI outputs are stably written into the file system, their nature changes.
Outputs are no longer disposable content. They become inheritable, modifiable, reusable work states.
Historical files record completed work. Active files carry ongoing progress. Goal files point toward intended destinations.
Work becomes a continuous trajectory rather than a pile of isolated results.
From Operational Momentum to Systemic Potential
Within this structure, systems begin to exhibit momentum and latent potential.
Work no longer depends on constant human intervention. It advances under established state and constraints.
Humans define goals and handle exceptions. AI executes within scope. File systems accumulate process and assets.
An “office that runs itself” does not emerge because AI replaces humans, but because work is placed inside a structure that both humans and AI can jointly advance.
Conclusion
From the internet era to the AI era, the center of product design is shifting from how information is presented to how productivity is organized.
When work is understood as continuous state progression, the core of product design is no longer entry points or interactions, but whether the system can carry that progression.
The file system is not a preference. Under current technical and cost constraints, it is a structural decision that makes human-AI collaboration viable.
What it defines is not a feature set, but a design judgment about whether AI can be absorbed into real productivity.
