Grant

It's kind of a creative solution because I wasn't sure how it was going to solve this problem on its own.

So it's like, okay, yeah, you just paste the inbox and sure.

Yeah, presumably so.

So let's say like you as a viewer are trying to do this and you come up with a great pitch and you put it together and it comes back with something that is not at all what you were thinking of.

There's two things you could do.

One, you can have it do what we're doing here, which is have it go back and rewrite everything.

But that's probably not practical.

Actually, what would be better is to say, OK, this is what happens when I don't give it all of the context it needs to succeed, which is where the term context engineering comes into play.

And instead, what I would do is I would spin up another instance and I would try it again, this time being even more specific with the goals that you're trying to accomplish.

Yeah.

And so if we were going to structure this prompt in a way that would probably be best for it to be received, we would want to give it the goal at the beginning, which the goal is to create this.

You don't really need to give it a role.

I think that's probably handled by this UI that Gmail's made or Google's made.

It knows it's an app-building agent at this point.

Yeah.

So you want to give it the goal, and then you'd also want to give it explicitly how you think of it as going to work.

Basically, you need what's called a...

like, well, I'm just going to simplify it, a vision for how it works.

So you need to have like a mental model of how you want this app to work and how you conceive of it working.

And you can break that out either, you know, very, very granularly, or you could just kind of broadly break it out, see what it comes up with, and then adjust your plan from there.