Unpacking the Argument
The author's analogy is databases and electricity: no one buys "a database" itself; people buy the accounting software and booking systems that databases support. The same goes for AI—the model is the underlying capability, and what users ultimately want is a concrete result like "finish my expense report for me" or "review my code for me." The flaw of a whole crop of "wrapper" products isn't the wrapping, but that the wrapper contains no deep understanding of a specific problem—the moment the model upgrades, their reason for existing gets eaten.
Look at the Products That Survived
This argument isn't new, but every time it's raised there's a ready-made comparison object. Look at the AI products that survived—almost all won on workflow rather than the model: coding assistants won on integration with the editor and codebase, and customer-service products won on connecting through to the ticketing system. There's actually a very handy filter for selection—ask "if the underlying model were the same for everyone, what would this product have left?" The answer usually makes the ranking obvious at a glance.
via: Hacker News