Carl Yeh

So nothing is always set exactly the same way.

So...

Really, again, back to what and why do we do things and and like when do we do it?

So we give this discount to this company in December, always in December.

You can add that to the agent workflow.

But let's say, oh, the CEO at that time said, you know what, we got to give them a discount because they'll probably pay more in two quarters, right?

how does your agent know that?

And so those are small examples of like the agent wouldn't be able to do that autonomously because they wouldn't have, you'd have to build something that takes in the count of all these decision traces.

And so hence at the end, I thought,

there's always going to be room for the human because only they can provide the context unless you can extract that decision trace.

I don't know how you'd extract that decision traces because that's how decisions are made unless you completely give it to the agent and it's making the decisions.

But then your question is, how is it making decisions?

What's it based out of?

Which means, is it based on your previous decision traces and so on?

So there's a big debate on that.

And there's apparently a big opportunity to extract those decision traces if SaaS companies can do that or the next level of SaaS, that's where it is.

No.

And then there is a post by...

Jaya's partner, I forgot what his name was, his business partner had to post on why Salesforce can't do that.

Because we saw, I don't know, a couple of days ago or a week ago saying, hey, Salesforce went back because LLMs can't be trusted, right?