[kylemacomber]

Don't tell anyone, but once you're on the paid plan we're not actually enforcing the 100 message limit right now ;)

We're still in the learning phase, and are going to adjust the plans based on exactly the kind of considerations you're raising

[markden]

Your secret is safe, hah! It's almost like you need an intermediate 'cheap/dumb' AI as a proxy to flesh everything out ahead of sending it off to be coded ... something that crisps up all the requirement and ultimately crafts a more cost-effective and likely all-around-better prompt. Even being semi-technical, I'm always surprised how much the details matter when describing something to AI (e.g., the fireworks should only explode _outward_, not inward, lol). Thanks!

[kylemacomber]

There's a lot I want to do to improve the prompting experience. As I mentioned in the original post, it's very simple right now.

There's a lot of inspiration we could (and probably will) draw from other products (e.g. Cline's Plan vs Act modes) to build what you're describing.

It can also be really fun and productive to just rapidly iterate on what the AI gives back, without having to take the time to describe it all up front. Sometimes this approach can lead to a new directions you might not have thought of.