Why is working in pairs so rare at work?

I agree that on-the-job working in pairs is rare. There are some exceptions such as pair programming and law enforcement (police work in pairs usually) but not many others!

Why are pairs so rare?

I attribute it to the fact that some supervisors and project managers either:

1. don’t know what is technically involved in enough detail to think of a way to allocate work to teams of two. When they do, it usually isn’t two workers who are peers. Instead, one is experienced and the other is less so or maybe new to the company or project, and mostly “shadows” the experienced person. It is more like one person working, and the other person watching and learning!
2. are concerned that two people who are peers—and should contribute equally—will not. One person will do most or all of the work. Why does this happen, and why doesn’t the one who contributes say something about it to management or the team leader? Lots of reasons, some of which I couldn’t ever figure out! Even when I suspected this was going on, it wasn’t possible to prove without the cooperation of the person who was doing the bulk of the work.

People can work in pairs most effectively when there are not multiple dependencies in the workflow/ timing. For example, one person might write the code to do some analysis, after or maybe simultaneously with the other person who looks for a data source then tests/confirms its adequacy (e.g. quality, frequency, time span, availability/cost, has a data dictionary) for the analysis. Yet even this scenario is more accurately described as two people working on two different parts of a project (or assignment, or experiment) rather than working as a pair.

An eponymous example

The only example that quickly comes to my mind is pairs programming. I did that at work, briefly. It was fascinating and revealing!

Neither of us knew Python, but he understood bash and Jupyter Notebook. He was a new hire, PhD Applied Math, formerly at NASA mission control, and had no subject matter expertise. I had subject matter expertise, no PhD, understood the data structures we would use, and knew our company’s systems and resource constraints. 

I was surprised at how often and quickly he could write logical queries that worked correctly, the very first time! I was also surprised at how often and quickly he would resort to searching on StackOverflow when there was anything that neither of us knew or could figure out together in less than a few minutes, or ever asking for additional context/ insight from others on our team.

I liked it, but he had enough after a month. (We did sessions twice a week.) He preferred to write most of the code for the statistical analysis on his own, then hand it off to me to describe our assumptions, determine what data should be excluded, do sanity checks on the output, etc.

At month end, he had learned Python but almost nothing about the data and subject matter of his new employer. I learned some Python, but didn't retain it. I mostly just learned how to make minor modifications to other’s code. I also got familiar with Jupyter notebook and liked it.

*The question didn’t specify the field of work. I tried to make my answer as general as I could.