A rational conversation on where AI is actually going | Benedict Evans

AI matches internet/mobile impact, not industrial revolution scale

Evans argues AI is fundamentally transformative like smartphones or the web, but not categorically larger, placing us currently in a "1997 internet" phase where most applications remain unbuilt and unreliable.

Massive adoption gap between tech insiders and general public

While some tech workers have abandoned Google for AI clusters, survey data shows even among teenagers, only 15-20% are daily active users, with 60% not using AI tools at all.

Historical pattern suggests job transformation, not elimination

Evans counters apocalyptic job loss predictions by noting that automation historically eliminates specific tasks while creating new roles that didn't previously exist, evidenced by AI labs themselves increasing headcount.

The hard part of work is judgment, not task execution

Using the spreadsheet analogy, Evans explains that automating tasks (like VisiCalc for accountants or IDEs for developers) typically increases output rather than reducing employment, because the valuable work lies in deciding what to build or analyze, not the execution.

Anti-AI sentiment offers moral superiority but no practical advantage

Dismissing AI as evil may feel satisfying, but Evans argues that practical success requires diving in to understand what the tools can actually do today.

AI labs investing heavily in human consultants

Despite predictions that AI would replace consultants, leading labs like OpenAI and Anthropic are acquiring professional services firms to help enterprises integrate AI, recognizing that implementation requires dedicated project teams.

Companies lack internal capacity for workflow redesign

Organizations have no spare staff to analyze which workflows can be automated or to integrate vertical systems, necessitating external consultants like Accenture or Bain to manage the transformation.