Today's GPT wrapper could be tomorrow's Salesforce

Andrew Chen had just put the skeptical case to Ben Horowitz, the venture capitalist. A lot of AI apps, Chen said, have little in the way of network effects or durable advantages, and critics dismiss them as GPT wrappers, meaning apps that put an interface on someone else's AI model.

"The GPT wrapper thing was always a bit of a dumb idea ... not to build one, but to call it that," Horowitz answered.

Horowitz argued he had heard the same dismissal before. When relational databases arrived, with their plain language for telling a database what to do, he said, people asked why anyone would need business software companies like SAP, which sell software for accounting, sales, and customer tracking, or, to use Horowitz's words, "in those days, Siebel or Salesforce or whatever?" That looked ridiculous to skeptics, he said, just a wrapper around the database.

That judgment, he argued, proved mistaken, and the same thing is happening with apps built on large language models. Those apps, he said, are pretty sophisticated models and applications on top of the foundation model. The coding assistant Cursor, for example, now runs about 14 models in a fairly sophisticated system on top.

That work gives the company data and an advantage others lack, Horowitz said. In his view, talking to, in Horowitz's words, "super high-end programmers all day about how to design their program" puts a lot of information into its models that nobody else has.

Then Horowitz added, "possession is nine-tenths of the law" in software. He asked how much better ChatGPT is than Grok right now, and answered that, starting from scratch, the difference is marginal. It is not a network effect yet, meaning a product that gets more useful as more people use it, just familiarity with something as simple as talking to it, but people tend to keep using what they use.

So taking a market requires much more than parity, Horowitz argued. As Horowitz put it, "you don't take out a competitor with an equally good product or a slightly better." You need one that is ten times better.