Bryan McAnulty

So after it edits a piece of code,

The system shows that back to it and then it decides, okay, now this is the next thing I'm going to do.

This is the next thing.

And some of these tools will even do things where like if there's a to-do list, like the agent has to, is forced by the programming to repeat until the to-do list is finished.

So that's the idea of a loop.

And some of the ways that you can do these things like inside ChatGPT directly or inside these Asian tools is a lot of them have like an automation section.

This is the easiest way to set something up as a non-programmer.

If you can think of a task that you would have repeated, you can have a small loop that is repeating daily, weekly, hourly for that.

So something could be

like find one small bug in my software or something and try to fix it or find one file that's getting too long in my software and try to optimize it and make it shorter.

And like, these are little things that maybe you'd want to spend some time on.

But like,

now the AI can do it.

You can just have that kind of running on a repeat process and it's just constantly approving.

It doesn't need your direct input necessarily.

Completely.

Yeah.

Yeah.

So with the combination of like the agent harnesses, a tool like latch loop, quad code, codex, and the model is getting better.

Now we're at the point that the model can continue towards this goal without having to use a continue, continue or OK, OK.