Andreas Welsch

It's very, very hard to optimize all three.

I would say it's impossible.

So yes, price is certainly one part, but how do you ensure that the output that is generated is relevant, is of high quality, is something that's usable, is something that you would

confidently give to your end user or to your end customer.

I think there we need to spend a lot of time and diligence, and I'm actually seeing large enterprises go in that direction and think about the governance.

How do we make sure that agents, for example, only have access to the systems and data that they need to have access to?

So we don't expose data that is not intended for, you know, this person maybe in a call center or this customer to see.

So quality, security, super, super important.

It multiplies the risks if we're not doing that right.

Otherwise, then it's just bad decisions faster.

So I've recently had some really, really insightful conversations with security experts, cybersecurity experts, and I'd recommend anybody to check out the OWASP top 10 for large language models.

I think they're working on something for agentic AI as well.

OWASP is an independent organization.

And, you know, seeing what some of those recommendations are that security professionals make and give you, it's very, very insightful because you also learn what are the things that could go wrong.

So how can we mitigate for those from the get-go as we develop systems, as we architect

new applications, how do you make sure that, for example, principles like zero trust are embedded from the very beginning?

We don't trust any piece of the architecture more than we need to at all.

So same with limiting access to data on a need-to-know basis.

What do you really need as an agent to do that task?

And some of those core principles that the domain has developed over the last 20, 30 years of the internet, cloud, and so on, now with AI and energetic AI, to me, that's a very important starting point.