Carl Yeh

So the thought, the piece was, how do you build such system when everything is still floating in all these different places?

And so he was saying that the opportunity isn't about the system of record.

It's actually at the decision traces level is how do you collect all those decisions, everything that goes into making a process.

So the example he was saying was like, okay,

there was a, when you ask about, hey, what's the ARR?

Well, like annual recurring revenue of a company.

When you talk to sales, they'll have a different answer.

You talk to finance, they'll have a different answer.

When you talk to marketing, they'll have a different answer.

When you talk to operations, they'll have a different answer.

And so like, first of all, you're like, hey, what is the level set of actually getting that answer?

But also that ARR could mean different things because like, hey, we gave a deal to this company because we always give a deal to this company because they're a larger company.

But that is never actually mentioned in the standard operating procedure or the actual how to do things.

So your agent wouldn't know how to do that unless you fed it.

And because it has no memory or the memory is poor, it has to remember you always have to keep feeding that.

The only thing that I argue against this is that that's great for historical use cases.

So historically, we gave a discount to this company every single year.

Well, how about if another company you give a discount to for a new instance?

I think you can only remember the past.

It can't actually take into account the future or current when the environment changes or something happens, right?