Andreas Welsch

Is it just processing?

Are they

wrangling the data and writing it back?

Are they making decisions?

And then the third one is then on a technical level, right?

We've seen this with prompt engineering in guardrails, for example, you're only allowed to do this, don't do that.

Or if you receive a question to which you don't have the answer, be honest, say that, you know, that's not part of your scope or you cannot answer this or point to a different resource.

So foundational, people, enable, training type things.

Two, security.

And then number three, additional technical mitigations as you're using these tools.

And you know, I'm still somewhat on the fence there.

If language, natural language is really the best way to give instructions.

So I would say for a good reason, right?

We've had programming languages for decades.

If this happens, then do that.

And it's very clear and very explicit.

If your program runs into an error, you can trace it.

You can say, okay,

you know, this variable has this value, hence it doesn't move forward or it throws an exception.

Now, if we move to agents and agents give other agents instructions based on natural language, I mean, we've all played Whisper down the lane.