Andreas Welsch

Yeah.

You start with one thing, you end up with something totally different.

I think that is a real and not to be underestimated risk.

As more and more agents converse with each other, check each other's output, give feedback and so on.

And coming back to security, right?

Even a malintent of one agent instructing another one to do something that they're not supposed to, right?

Or go off the rails or, well, you know, it's perfectly fine.

This is an exception.

This is just one time.

So, you know, just thinking about the social engineering of people applied to technology, I think is something that we just need to have on the radar as well.

And just think about as we build these architectures and systems.

So intelligence is just one factor of your equation, I would say.

Whatever service or product you provide, there's a whole lot more to it, I would hope, than a wrapper around an LLM.

If that's your business model, you might be in trouble with the next upgrade you want to spend or make, right?

We've seen this with PDF wrappers and chat with your PDF type things, right?

And all of a sudden it's a standard functionality and one of the tools we use on a regular basis.

So

Please don't build an LLM wrapper.

You need to have some moat of your own and some secret sauce and expertise.

And I think it's that secret sauce and expertise that makes your business unique.