Coding is AI's first undeniable use case, and Benedict Evans uses that fact as a lens to examine everything else. In a conversation with Erik Torenberg, Evans argues that coding works because it has a tight feedback loop, measurable output, and a user base already fluent in iteration. That specificity matters. Most AI use case debates stay abstract. This one does not.
The conversation gets sharper when Evans applies platform history to today's pricing crunch and CapEx problem. He draws comparisons to earlier infrastructure cycles to ask whether foundation models capture value or become commodity pipes. The OpenAI versus Anthropic strategy section, starting at 5:53, is worth close attention. Evans frames their divergence not as branding but as a structural bet on where margin lives in the stack.
The unresolved questions are the reason to watch the full video. Evans is direct that enterprise adoption, consumer behavior, and model commoditization are all still open. He does not pretend the trajectory is clear. The section on what comes after coding, at 22:48, is where the analysis becomes genuinely speculative in a useful way. Evans earns the uncertainty rather than hiding it.
[WATCH ON YOUTUBE →]