Benedict Evans 拆解 AI 炒作周期：历史类比、工作与价值归属

Source-faithful non-verbatim point: Saying AI is like electricity or bigger than the internet does not explain the mechanisms through which economic change will occur.

Source-faithful non-verbatim point: Unlike earlier platform shifts, AI leaves us uncertain about the physical and scientific limits of the core capability itself.

Source-faithful non-verbatim point: Because model capabilities are jagged across tasks, adoption and economic value will also appear unevenly across jobs and industries.

Source-faithful non-verbatim point: Coding became an early enterprise use case because outputs can be tested and verification can scale alongside generation.

Source-faithful non-verbatim point: Historical automation frequently increased the volume and scope of work instead of simply eliminating the occupation it touched.

Source-faithful non-verbatim point: Most valuable AI use cases will need entrepreneurs to invent products and workflows rather than waiting for spontaneous user behavior.

Source-faithful non-verbatim point: Moving from isolated pilots to structural enterprise change is a difficult transformation project that often requires outside expertise.

Source-faithful non-verbatim point: Foundation-model labs may resemble TSMC more than Windows: extremely valuable suppliers without necessarily owning every downstream layer.

Source-faithful non-verbatim point: Most organizations have urgent operational, regulatory, and physical priorities that can outrank AI adoption.

Source-faithful non-verbatim point: The durable significance of AI lies in its enabling capability across products and processes, not only in visible chatbots, text, or images.