Nate Meyvis

Why use a generative AI browser?

On a recent episode of The Talk Show, John Gruber suggested that generative AI browsers, most notably ChatGPT Atlas, lack use cases.

I'm sympathetic to Gruber's take: it is much easier to see why usage of ChatGPT Atlas is good for ChatGPT than it is to see why it's good for the user. But here are a few use cases:

  1. Browsing a forum that requires a login. ("Please browse around for a while and summarize all the most credible rumors about X.")
  2. Shopping. (I don't do this, but Joanna Stern does.)
  3. Helping me migrate my blog to Bear.

What do these things have in common?

  1. They're predictable enough that AI is good at them.
  2. They're not easily scriptable, because API access isn't available and/or they require a bit of judgment.
  3. It's pretty easy to catch and fix bad outputs.
  4. Costly mistakes are either impossible or easy to avoid.

There's an argument to be made that this category of task might go away. Take shopping: online grocery stores might expose tools that allow you to shop with generative AI without using a browser, or third parties might develop new AI-assisted shopping tools. (Some of this is already happening.)

That said, a lot of us do things on the Web that are time-consuming and don't require that much good judgment. As long as this remains true, there's at least an argument for being comfortable using an AI-first browser.

#generative AI #productivity