Jyunmi Eran

One thing that I just want to mention on that is...

The hardest part of, like, when you're trying to look at an MCP and what it wants to do, and especially if, like, we're trying to custom build some, like, what are the tooling you want it to do?

Especially, so, for example, it's easy for things like an Asana or things that people use, but...

It gets harder when you're trying to do proprietary or industry-specific software that only an industry uses, like a specific construction.

So to build an MCP on top of that is very difficult.

And sometimes it's near impossible just because of how the software is created.

And it was never created to have an MCP on top of it.

And their APIs are

it costs to use.

And then once you see some MCPs, it's like, holy smokes, you have 75 different tools that the model has to review and pick which one.

So I think the...

I think the workaround was that is instead of creating an MCP that allows an AI model to do all the tooling, to look at all the tools, you create an MCP that has specific processes, right?

So you bundle those tools.

So it's like pick which process you want it to do rather than pick the tool that

and have you do it.

So it's sort of, you, you do, you do calm.

One could say combos of tools that the model would pick or combos of use cases based on the tools, rather than just pick a bunch of tools and then figure out the use case from there.

Um, granted that was, I learned that from, um,

The Australian AI show that they call themselves.

Yeah, the red logo.