It's not a "business problem, organizational problem, and communication problem". (Or not always: caveat at the bottom.)
Conway's Law isn't a flaw, it's the correct way to engineer a product, whether the organization is dysfunctional or functional.
Humans working together on the same product need to communicate. The farther away they are, the more this communication becomes harder, fragile, and prone to misunderstandings.
Thus, the boundaries between their areas of responsibility must be manageable via whatever communication resources are available.
If two people are working side-by-side and constantly aware of each other's work, they can easily cooperate on a single imperative function with no isolation whatsoever.
If they're totally separate teams that normally communicate via emails and need to schedule a meeting a week in advance, then you need to have iron-clad API contracts between their codebases to enable them to work in peace. The other option is to get the teams closer, but that may simply not be possible past a certain scale - you just can't get 100 engineers to know what everybody else is doing.
Sure, if you have a team of 8 people and they have no idea what each other is doing because there are 16 middle managers wasting their time? Then yeah, you absolutely have a business problem, and Conway's Law will be a bandaid over it - you'll split the product into 8 libraries or 8 microservices, wasting a bunch of resources but keeping your engineers sane. But if you solve the problem and put the 8 engineers in a shared room so they can collaborate efficiently, and get rid of the no-longer-necessary boundaries, you're still following Conway's Law.
The thing is, it's not written in blood. You don't have to reorg to redefine your architecture. You don't have to design an ironclad API, or work on an "imperative function" between people sitting in the same room. In fact, there may not be an organizational principle that will give you a good architecture.
Want evidence? Open Source software. People sitting halfway around the world making software that has eaten the entire planet. Some of the contributors have commercial interest, but they don't get to dictate the architecture. And hey, how about that, the software gets made well anyway, even though the communication structure is often just one global mailing list. "The org structure" doesn't define the architecture because it would become random. Instead, the software's architecture is defined, and then people go out of their way to figure out how to make it. It is the opposite of Conway's Law. Yet it has been working for over 30 years.
Conway's Law is a flaw in human ingenuity, like a bias. We have many biases. Bias is an emergent property that, if unchecked, leads to bad outcomes. But we are capable of rising above them, if we try. The result is better if we do.
Written specifically to match the politics of Linux kernel development, since there wasn’t an open source alternative that could do so already (paraphrasing the interview)
Git is a great example of terrible design; Linus has said this repeatedly (and it's in the name).
Most successful open source software is simply designed however one person thought it made sense. Often this design ends up being reworked into a new major version, but again, in most open source projects the goal is just better design without deference to communication structure.
Conway's Law isn't a flaw, it's the correct way to engineer a product, whether the organization is dysfunctional or functional.
Humans working together on the same product need to communicate. The farther away they are, the more this communication becomes harder, fragile, and prone to misunderstandings.
Thus, the boundaries between their areas of responsibility must be manageable via whatever communication resources are available.
If two people are working side-by-side and constantly aware of each other's work, they can easily cooperate on a single imperative function with no isolation whatsoever.
If they're totally separate teams that normally communicate via emails and need to schedule a meeting a week in advance, then you need to have iron-clad API contracts between their codebases to enable them to work in peace. The other option is to get the teams closer, but that may simply not be possible past a certain scale - you just can't get 100 engineers to know what everybody else is doing.
Sure, if you have a team of 8 people and they have no idea what each other is doing because there are 16 middle managers wasting their time? Then yeah, you absolutely have a business problem, and Conway's Law will be a bandaid over it - you'll split the product into 8 libraries or 8 microservices, wasting a bunch of resources but keeping your engineers sane. But if you solve the problem and put the 8 engineers in a shared room so they can collaborate efficiently, and get rid of the no-longer-necessary boundaries, you're still following Conway's Law.