We can build a three-layer framework for understanding the AI era:
Layer 1: the technology stack (bottom to top)
Bottom layer: model capabilities (Pre-training → Reasoning → Long Horizon)
Middle layer: infrastructure (Framework → Harness → Memory System)
Top layer: application scenarios (Coding → SRE → Research → Vertical)
Layer 2: the value chain (from supply to demand)
Supply side: AI creates new capabilities (for example, a Long Horizon Agent)
Transmission layer: new capabilities reshape workflows (for example, GitHub changing from a repository into a task assignment system)
Demand side: new workflows create new demand (for example, “using AI to write code” becoming the norm)
Layer 3: the competitive landscape (from general to specialized)
General layer: model vendors (OpenAI/Anthropic) compete for the entry point
Infrastructure layer: independent players (LangChain/MemOS) provide portable capabilities
Application layer: vertical companies (Factory/Rogo) build moats in specific scenarios
The core logic of this three-layer framework is:
The technology stack determines what can be done: the breakthrough of Long Horizon Agent comes from the combination of model + Harness + Memory
The value chain determines how value flows: breakthroughs on the supply side create new demand, rather than satisfying old demand
The competitive landscape determines who can win: the general layer is winner-takes-all, the infrastructure layer allows diverse coexistence, and the application layer iterates quickly
For different roles, the strategic meaning of this framework is different:
For founders:
If you are working on the bottom layer (model/engine/memory system): pursue generality, build technological barriers, and aim to become infrastructure
If you are working on the middle layer (Harness/toolset): pursue best practices, iterate quickly, and aim to become the de facto standard
If you are working on the top layer (vertical applications): pursue scenario specificity, accumulate memory assets, and aim to build a moat in your domain
For investors:
Look at the bottom layer: look at technical barriers, and whether there is a possibility of “countering” other technologies
Look at the middle layer: look at engineering capability, and whether model capabilities can be turned into usable products
Look at the top layer: look at scenario depth, and whether an unreplicable memory asset can be built in a vertical field
For individuals:
At the cognitive level: switch from the “demand side” to the “supply side,” from “what do I need” to “what can I create”
At the capability level: Coding may become the new “literacy.” You don’t necessarily need to become a programmer, but you should understand code logic
At the asset level: the collaborative memory between you and AI is your personal asset, and you should manage and accumulate it consciously
For organizations:
At the strategic level: determine whether you are “working for AI” or “being transformed by AI” — embrace the former, be wary of the latter
At the organizational level: a memory system is not only a technical issue, but also an organizational issue; it requires designing permissions, isolation, and sharing mechanisms
At the cultural level: from “people using software” to “AI coordinating people,” organizational culture needs to adapt to a new mode of collaboration