Corey Knowles

What are your questions?

Yeah.

So I think one thing is it's like a lot of the code that we tend to be writing, you know, it's more like if you want to use the term, it's like more like vibe engineering.

Right.

So like people are people like reading the code and like looking at it.

And so unless we're just prototyping or something, usually the workflow is much more human in the loop than just go off on a prompt.

So I think oftentimes you'll see things where people will ask for a plan and then they'll work on the plan.

like together or maybe you'll ask for the agent to just like build the entire feature that you have in mind actually you know we we tend to like doing something we call best event where like in codex cloud we'll like run codex like maybe say four times on the same idea same task or even locally we might do it in work trees or like multiple repo clones

I can show that, but I think it's a bit advanced.

Maybe I won't show it.

But the point is that you'll ask the model to solve the same problem four times.

And then we might look at it and be like, oh, wow, I didn't even think about doing it that way.

Or now I understand that there's this like other way to do it.

And then then I might just like start from scratch and be like, cool.

OK, here's exactly how I want you to do it.

Right.

Because we've looked at a plan or we've seen something and like do often like and then like create aggressively create new new chats if you find the agent going getting stuck.

and ask for like very specifically what you want.

Personally, I have found that I don't need to create new chats nearly as often as I used to.

Like I used to create new chats really aggressively because I would find that the model would like not do super well as context got really long.