The long version of why it exists:
Historically, a Python package has been defined by the existence of the setup.py file. Executing this file with various command-line arguments is how packaging tasks are carried out: 'setup.py install' installs the package, 'setup.py sdist' builds a source distribution (.tar.gz) suitable for uploading to a package repository, etc., etc.
But this workflow has dependencies! Most packages will only depend on setuptools (which supplies the setup() function) and wheel (which supports building binary packages), so the packaging tooling assumes the presence of setuptools and wheel and you never have to specify them as explicit dependencies (in the old old days, setuptools used a drive-by installer file bundled into the package so people who didn't have setuptools could still install setuptools-based packages; nowadays, pip supplies it automatically).
But what if you have a workflow that requires something else? What if you use an alternative package manager, or package build tool, or really anything that isn't setuptools or wheel? How do you specify the toolchain you need to work with your package?
That is the thing pyproject.toml solves. Since it's not an executable Python file, it can't get stuck in a chicken-and-egg problem ("execute setup.py to find out what you need to execute setup.py"), and it can specify the necessary build system for a package in a standardized way.
pyproject.toml is not the place where you declare that dependency. This is why I am pushing back on you, because your comments all very very very strongly imply that you think pyproject.toml is a place where you would declare that dependency.
pyproject.toml is where you'd declare that you're using an alternate build/packaging toolchain. You would then use whatever mechanism that toolchain provides for specifying your dependencies (in the case of the default setuptools build/packaging toolchain: install_requires in setup.py).
https://github.com/sdispater/poetry/blob/master/pyproject.to...
But you've caught on to the key thing, which is that neither they, nor the default setuptools-based toolchain, have to do that. The pyproject.toml file exists to solve the chicken-and-egg problem of specifying that you use an alternative toolchain, when figuring that out might otherwise be dependent on your preferred toolchain already being present and configured.
The reason for the clichéd build you up, tear you down, repetition process throughout all of human history, is that a vast number of people will be rage monsters, petty, filled with envy and jealousy. The types that are prone to that ill behavior, will come pouring out of the woodword once you've acquired any consequential amount of success/money/fame/whatnot. These are not good people, they are not capable of reasoned discourse, their opinions do not matter, their hostile negativity should be ignored and disregarded accordingly (they're also, fortunately, extraordinarily easy to spot and separate out from the more reasoned criticism; their entirely unnecessary over-the-top negative behavior provides for easy filtering).
I'm more upset about the whole pyproject.toml thing that means I have to migrate everything again, from Pipfile this time, even though I have barely finished migrating to it!