Corey Knowles

It's 4% of an employee's time.

But it's not, you know, there's no single task that it can do end-to-end, not a single job it can do end-to-end.

Yeah, I can do this one part in the middle, but you're like, okay, yeah, keep one on the side, keep one on the side.

So, you know, is it really safe?

Like, does, if you're Procter & Gamble, are you selling more Tide Pods because you bought ChatGPT?

Definitely not.

Are you saving money on the bottom line?

Maybe at the margin, but it's not like you can hire that many fewer people.

It's not like tasks are being automated wholesale.

And so that's the problem with chat, I think.

And so that's where I think agents are really, this is why I think there's so much interest in agents, is can we try to get a whole task automated end-to-end?

And this, I think, is the magic behind something like Cloud Code or Cursor, which is

if you think about it, you could actually do what Cloud Code does in ChatGPT.

You know, you could, instead of having Cloud Code run a for loop, you could just have the engineer do it.

So you could say to the engineer, ask ChatGPT, you know, write me this code.

Let me copy it, paste in my ID, press run.

Okay, it doesn't work.

Let me paste the error message back to ChatGPT.

Okay.

I've done a lot of that.