Part X: Ecosystem and Economics

"A wealth of information creates a poverty of attention."

Herbert A. Simon, "Designing Organizations for an Information-Rich World"

The previous parts have shown how models are trained, served, composed, measured, and constrained. Part X asks who can afford to do those things, who packages them for everyone else, and how the prices feed back into design. The ecosystem is not scenery around the technical stack. It is one of the forces that decides which parts of the stack become common practice.

A model release is an economic object as much as a technical one, and Chapter 73 reads it on both axes: weights or API (open-weight models you run yourself versus closed models behind an API, a distinction that section defines), open source AI or open weights, capability or efficiency, license or lock-in. Chapter 74 then treats the model as a file in a software supply chain: serialization formats and why one of them runs code, the hub as the field's registry, and the poisoning and backdoor attacks that no byte-level check can catch. Chapter 75 follows the standards and layers that grew around model calls, tools, agents, gateways, sandboxes, observability, and authorization. Beneath both lies a cost structure, which Chapter 76 lays out: training and inference run on different clocks, and market prices move as the field matures.

The second half of the part turns from artifacts to markets. Chapter 77 maps the AI value chain from silicon and cloud to model APIs, applications, and data. Capability does not automatically become productivity, and Chapter 78 asks when it does so in a real workflow rather than a demo. Chapter 79 treats data as a rights-bearing supply chain, where provenance, consent, documentation, and compliance determine whether a corpus can become a product input. Chapter 80 closes the part where agents become economic actors: the identity, delegation, and payment rails that machine commerce is still building, and how much of that economy is real.

This part shifts attention away from any single leaderboard point and toward the terms under which capability is made available. A model can be strong and still unavailable, cheap and still costly to operate, open and still hard to use responsibly. Economics is where those differences become part of the architecture.