To be honest I fail to see how this is more compact and sleek.

With patterns one matches on the structure of the data. It’s explicit and describes exactly how the data should look like. Guards exist to enforce further constraints which cannot be easily expressed in a structural pattern, such as a value being an integer.

That said, your particular examples can be expressed using structural patterns (and I would argue more clearly):

def my_function([x1, x2] = list) do
  # ...
end

def my_function([x1, x2, x3] = list) do
  # ...
end

It mixes concerns.

On the one hand you have structural matching using patterns and on the other hand you have logical matching using conditions (guards). Both are useful but serve subtly different purposes.

Mixing them is not necessarily a bad thing but from where I’m standing it’s good to know that a pattern is a pattern and a guard is a guard. I can look at a function header and know that there is no logical component to the match (such as an or) as long as there isn’t a guard clause.