AI Eats the World: Benedict Evans on the Next Platform Shift

Key Takeaways

AI faces a significant gap between industry excitement and widespread daily use cases despite tools like ChatGPT having 800-900 million weekly active users.
The central debate revolves around whether AI represents another platform shift or a more fundamental technological change, impacting new company creation versus existing tech giants' strategies.
Major tech companies are reorienting their strategies, but AI's ultimate impact across various industries and its inherent physical limits remain largely uncertain.
Future AI development must overcome current bottlenecks in compute and determine if it will foster entirely new user behaviors and product categories beyond current applications.

The term "AI" often applies to new developments, distinguishing it from established machine learning technologies.
AGI (Artificial General Intelligence) is perceived as newer and potentially more impactful than current AI.
Technology analyst Benedict Evans compares AI to historical platform shifts like PCs and smartphones, noting some industries may be profoundly transformed while others are minimally affected.
Major tech companies, including Google, Meta, Amazon, and Apple, are actively repositioning their strategies in response to AI advancements.

The core debate questions if generative AI is a platform shift akin to mobile or a more fundamental technological change like computing or electricity.
A tension exists between claims of human-level AI capabilities and the API-driven software development that characterized past tech booms.
Unlike previous platform shifts, AI's physical limits are unknown due to a lack of theoretical understanding of its workings and human intelligence, leading to more subjective predictions.
This uncertainty contributes to significant investment, potentially leading to bubble-like market behavior, though predicting its timing or nature is difficult.

Current AI investment is heavy due to perceived threats from competitors, carrying a risk of overinvestment similar to past technology booms.
Discussions bifurcate between technical aspects like chips and data centers, and the business models for AI-powered SaaS companies.
Obvious generative AI use cases include software development, marketing, and enterprise solutions, alongside applications for flexible, knowledge-based jobs.
Examples include individuals in Silicon Valley optimizing workflows and companies improving content generation, while a significant portion of the population does not use tools like ChatGPT regularly.

A significant portion of the population, despite having access to AI tools like ChatGPT, is not using them regularly, raising questions about adoption barriers.
New products often leverage existing concepts like databases and CRMs to solve specific industry problems, leading to a proliferation of specialized SaaS applications.
The true potential of current AI, much like early computers and smartphones, lies in enabling entirely new capabilities rather than solely perfecting established tasks.
AI errors in factual tasks necessitate human oversight, making it less efficient than manual work in cases where precision is critical.

The discussion queries whether AI will create new user behaviors and new companies, or if existing model providers will dominate, using the analogy of mid-90s operating system shifts.
The conversation explores 'how far up the stack' AI models can go, likening it to how operating systems and applications abstracted lower-level functions.
Users typically purchase integrated solutions, such as Everlaw's legal discovery software, rather than building with raw AI tools.
Effective user interface design, informed by user research and institutional knowledge, offers a valuable contrast to raw AI prompts that require users to define tasks from scratch.

Raw chatbot interfaces are critiqued for demanding extensive user input, raising questions about whether they are true products or merely disguised chatbots.
The evolution of user interaction with new technologies, such as Google Maps and Instagram, historically takes time to manifest its full potential.
Foundational AI models may not encompass all future specializations, suggesting that multiple companies will likely succeed in various AI sub-sectors.
A disconnect exists between AI model benchmark scores and actual consumer usage, with factors like brand recognition and distribution potentially influencing market share.

Companies with large user bases but lacking infrastructure or network effects face fragility, necessitating rapid product building and infrastructure engagement with partners like NVIDIA and OpenAI.
Google's position with its Gemini model benefits from existing cash flow from other products to sustain the significant cost of frontier model development.
AI is considered transformative for Meta, potentially reshaping search, content, social media, and recommendation systems, necessitating their own models.
For Amazon, AI could significantly improve large-scale recommendation and discovery, addressing a weakness in its commodity retail model.

Apple's past Siri demo showcased advanced capabilities like multimodal, on-device, agentic actions that are still not reliably achieved by major tech players.
A core question for Apple is whether AI represents a fundamental shift in computing, which would be problematic if they don't lead, or merely another service, similar to Google Search.
The discussion draws a parallel to Microsoft's loss of the PC platform in the 2000s, questioning if AI could lead to a similar shift where Apple might lose its platform advantage.
A potential future where AI fundamentally changes software, eliminating traditional apps, could still benefit Apple if future AI-access devices resemble current smartphones.

AI is expected to unlock new revenue streams beyond simple automation, prompting a reevaluation of existing business models and potential market redefinition.
This redefinition potential is compared to the internet's impact on newspapers by unbundling core functions and creating new paradigms.
The conversation questions what specific, falsifiable criteria would need to be met for AI to be considered more impactful than the internet or personal computing.
Historical technological shifts, like the iPhone and the internet, suggest their true impact was not immediately apparent, and current AI capabilities are still limited, not yet a replacement for human interaction.