Jamie Taylor

Yeah, it could be.

It could be a vegan bacon.

It could be bacon.

It could be whatever.

Absolutely.

You push the button and something yummy comes out the other side, right?

That's the level you're thinking when you're doing the spec.

Then when you do the plan, you're thinking, when I push the button, it needs to call this API, which tells that person to go make me my breakfast thing and then to bring it to me, right?

So you're thinking it's slightly different levels.

And then what you need to be doing is you need to be watching what the LLM is doing.

And the more effort you put into spec and plan of observing and pointing in the right direction,

the less effort you need to put in when it comes to telling it to implement the tasks, right?

So there's this almost like a funnel of the amount of information you need to give it.

You need to give loads of information in the spec side and then the plan side.

And then when you get to implementation, you don't have to worry about it so much.

You do have to watch what it's writing.

I'm not saying just put it in YOLO mode and then submit the code, right?

So that's the first big limitation is that you really need to know and understand the problem you want to solve.

Yes.

The only downside to that is whenever I've told people about this kind of thing or any kind of AI MLM-based coding system is that