Constellation (and VMS in general) is more than Skin-Deep

Why Non-Functional Requirements (NFRs) make mission-critical software behaviorally harder to disrupt than you think

I empathize with how gnarly the AI conversation gets when thinking about its impact on Constellation Software (or any Vertical Market Software) I get it. The fear is real.

But coming from the inside, having worked on machine learning and specifically having been tasked with rewriting legacy software for an “AI” implementation, I can tell you that it is not trivial.

Let me explain this using a fundamental software engineering concept called Non-Functional Requirements (NFRs).

When analysts evaluate $CSU.to or $TOI.v, or even when you just pick up your phone, you usually only focus on the Functional Requirements. These are the user interfaces, the buttons you click, and what those buttons do. Most people don’t consider the “invisible” issues until something fails. You know the feeling when an app doesn’t launch, crashes, or is excruciatingly slow.

These invisible factors are the NFRs. They define the quality of the system, not just the features. Here are the broadest categories:

• Reliability

• Availability

• Scalability

• Testability

• Performance

Optimizing for any one of these is hard. Optimizing for all of them simultaneously requires massive trade-offs in compute and resources.

Now, look at current-day “AI” and the chatbots you interact with. Where do you think they struggle the most compared to legacy VMS systems?

You can pick almost any item from that list, and you would be right. Current AI models notoriously struggle with these specific requirements.

The bottom line is this:

For mission-critical software where users and systems are fault-intolerant, and where errors can be fatal or catastrophic, we cannot solely rely on today’s AI systems. And truthfully, we don’t know when AI will achieve the accuracy required to change that.