Interview
AI Eats the World: Benedict Evans on the Next Platform Shift
- Generative AI adoption is projected to reach 10–15% of the developed population daily and 20–30% weekly within a short timeframe, with immediate utility expected in software development, marketing, enterprise use cases, and roles with open-ended workflows.
- Current market dynamics suggest a high risk of an AI bubble similar to the 1997–1999 internet era, driven by hyperscalers prioritizing over-investment over downside risk and potential market saturation if demand for AI compute falls short of expectations.
- The trajectory toward Artificial General Intelligence (AGI) is characterized by uncertainty regarding physical limits and a pattern where AGI is either already present as software or consistently forecast to be five years away, with the definition fluidly shifting as technologies become reliable enough to be categorized as standard software.
- Significant structural changes are anticipated where new billion and trillion-dollar companies may emerge from platform shifts, though value distribution may depend on the creation of dedicated UIs and workflows rather than solely flowing to model providers.
- Major technology companies face specific strategic risks: Apple may lose its device layer dominance if computing shifts fundamentally to AI agents, Amazon faces potential disruption in recommendation and discovery models, and Google's search business is expected to absorb AI as a feature rather than being displaced.
- Forecasting fundamental AI capabilities is deemed impossible beyond the near term due to a lack of modeling equivalents, leading to reliance on "vibes-based forecasting" and the expectation that most current assumptions about AI use cases will be proven wrong in hindsight.
- The market will likely produce multiple winners in every category, including model providers, as the technology evolves from raw chatbots to specialized products requiring human validation mechanisms and curated institutional knowledge.
- A perceived disconnect exists regarding OpenAI's progress, specifically between claims of deploying PhD-level researchers and the current reality of building software that merely accelerates other software development, prompting OpenAI to build an ecosystem of products to create stickiness around a commodity model.
- While some industries will undergo complete transformation, others will only receive useful tools, with a significant portion of the workforce struggling to map AI to daily tasks until the technology is integrated into familiar product workflows similar to Excel or CRMs.
- Generative AI is expected to initially excel at tasks previously impossible rather than replacing "old things," with new use cases emerging over time similar to the evolution of smartphones, eventually requiring the technology to function as a "person" rather than a guarded tool to be considered bigger than the Internet.