Writing scientific papers is very much a collaborative effort. Even when you are a single author, you still usually have to go through peer review to see your manuscript published. At a minimum, you have to satisfy one or two reviewers and the journal editor that it is acceptable; in other cases, it might be three or four reviewers, a managing editor and the chief editor. And all of them may have ideas about how you have to change or rewrite some part of the manuscript.
If you have one or several co-authors it becomes even more difficult because they surely have even more to say about how exactly the manuscript should look like when their name is on it than an anonymous reviewer. How well the collaboration on a manuscript works out is very dependent on the compatibility of the coauthors' personalities.
One thing that I learned early, long before going into science, is that every text needs a lead author. If two or more people try to write it at the same time by dividing up the parts, things can go wrong very quickly. In the simplest case, you may end up with a manuscript that looks very inconsistently written. I once reviewed a paper from an otherwise highly competent research group where several paragraphs in the discussion stuck out like a sore thumb due to their significantly worse phrasing and grammar; one look at the author list gave me a pretty good idea who might have written that part. Later I read a manuscript whose discussion was clearly an incoherent collection of several independently written paragraphs, each of them representing an individual coauthors' hobby horse that just had to be inserted there whether it fit or not.
In other cases, you may find that one coauthor holds everything up because they do not deliver the part assigned to them, or one cannot even start because everybody is arguing about the structure. It is much more efficient if one person is tasked with writing a draft which can then serve as a basis for suggestions by the coauthors. (An obvious exception are very specific segments of the materials and methods section that only one of the coauthors can provide because the lead author is not an expert on that particular method or analysis.)
When we have progressed to the stage of having a draft in our hands, the real fun starts. Unsurprisingly, the "freelance English teacher" type of peer reviewer I described in a previous post also comes in the shape of a coauthor. In extreme cases, every draft you send them comes back with suggestions for "improvements" for every single sentence, even for sentences that are now exactly as they wanted them to be one draft earlier.
One thing that irritates me particularly is when a coauthor insists that a sentence describing something universally known and blatantly obvious needs a reference, on the lines of: "The sky is blue (need refs for that)". Without providing those references themselves, of course, because they know they must be out there but just can't remember where at the moment. But it is generally hard to find decent references for the obvious - it is not as if a good scientific journal would gain much from publishing a study merely confirming what everybody has known for a long time.
In general, I find it pretty annoying if a coauthor produces several paragraphs of text without specific references but comments like "refs" or "Smith et al. 199X, look that up" in the brackets. Instead of inserting the references themselves they simply drop it all back into your lap half finished and leave you to guess what papers they were vaguely thinking of when they wrote that.
Often such a situation arises from the collision of two different styles of writing. Some authors, including myself, prefer to invest a lot of time and effort into the first draft and then have less work later. Others prefer to just let their unordered thoughts flow onto the page even if that leads to a considerably more rearranging and rephrasing later. They also tend to leave a lot of ellipses that have to be filled in later, and those include references that they cannot be bothered to look up at the moment because it would interrupt their train of thought and make them forget what they wanted to write next. Cooperation between the first and the second type of author can be trying at times.
The most difficult situations, however, come from the disagreement of coauthors about the overall direction the manuscript should be going. It either delays the writing process considerably or seizes it up entirely. There are at least three potential sources of conflict: scope, methodology and interpretation.
In the first case, the authors disagree what the paper should even be about. I actually lived through such a case myself once when three of the four coauthors wanted to tell entirely different stories, one about phylogenetic relationships, one about methodological problems and one about evolutionary biology. In the second case, authors disagree about the best way to analyze the data. That sounds rather trivial but can be bitter if different schools of thought are involved and even more so when combined with the third case, a disagreement about what the data mean.
The root of the last type of conflict generally lies in the irrationality and confirmation bias exhibited by at least one of the authors who is unwilling to accept that the results disprove an idea that they have publicly promoted. They may then try to twist the conclusions to be at odds with the actual results, or they advocate the use of clearly inadequate analyses because those would just fail to produce the relevant P value (or whatever). Luckily, I have never been directly involved in such a conflict but I have heard of one or two such cases among colleagues I know.
Again, individual scientists can be as irrational as anybody else, but because we constantly test and criticize each other we collectively get it right in the end.