In a lot of Star Trek episodes, the Engineer character has to do work like "modify the deflector array" so that it can work with the Romulans' ships or whatever. The Enterprise can't just automatically do that on its own. That's what I'm talking about here.

So not only does this hypothetical AI agent need to know *how* to operate the novel system, the novel system also needs to be able to be *operated upon* by the AI agent. This gets extraordinarily complicated once hardware starts getting involved in the situation as well.

But what I'm essentially saying is, we actually need something like a universal API that all software can interface with. Otherwise, even a very good agentic AI system is limited by how well it "understands" another (novel) external system.

In order to even support something like that right now, *someone* has to write a code translation layer *somewhere* that makes it so Alexa knows how to turn on a specific brand of dishwasher. And sure, dishwasher manufacturers can try to agree on a standard set of APIs for operating them digitally.

Without this missing feature of software engineering, you will never be able to, say, tell Alexa to turn on your dishwasher that it knows nothing about.

"Seamlessly" is one part of it, but another super key piece is that software needs to be able to communicate with other software that it's never encountered before. I can't even begin to guess how many decades away current genAI and agentic AI stuff is from that capability.