It never works, of course. Insights come from individuals, not collectives, and while having similarly skilled colleagues to bounce ideas off of and refine them with works well for great programmers, they have to come up with the ideas themselves in the first place. The communication overhead in a team of good programmers prevents them from having great-programmer insights at great-programmer speeds. The way to fight communication overhead is to ensure that each team member has a large corpus of shared information. In other words, a shared, standardized coding style which further hampers insight because insightful solutions have little regard for "the rules".
But most companies just want CRUD+business rules, so RAID (Redundant Array of Intransigent Developers) works well.
Pair- programming is a case in point. It emerged from "Xtreme Programming" which emphasized "values" and "people", not problem solving skills. So why didn't they go further and require "triple programming", then "quatro programming" and so on?
The way to tackle complex problems is by divide-and-conquer: Divide the problem into smaller parts which can be solved independently. Pair-programming is not that, it doesn't divide the problem into two and then let each programmer work on one part of it individually. Instead there's a pair of chefs in the same kitchen team-cooking the same spaghetti.
To put it bluntly I think people who demand "team-players" are typically those who don't have the skills to solve any part of the problem by themselves. By building a team around them they can hide that fact. If the project fails it is nobody's fault, the whole team did their best and still it couldn't succeed. We followed the best agile practices you know. Or maybe it succeeds, because of a few good men and women.
(Successful) team-playing is when each team-member plays their part, not when they all play the same part. Think volley-ball, that's a great team-sport.
Article on them here: https://www.wired.com/2012/08/google-as-xerox-parc/
I can see that pair-programming can help in some situations like getting new team-members up to speed. But I question those who say it should be applied universally, always, as a general rule.
Bouncing insights off each other in a team might lead to someone else taking up and working on something that to the originator was just a fleeting side idea in a train of thought, but can turn out to be valid on its on / viewed from a slightly different angle / in a slightly different context. Whereas if the individual works entirely on their own, valid ideas / insights may be forgotten and fall by the wayside forever.
(Personally I probably tend more towards the solo side, but have to admit there are advantages to teamwork too.)
With this in mind it makes sense to hire team players who are good enough over highly skilled workers who work in isolation. The trick then becomes finding highly skilled workers who are able to engineer and manage solutions for the lower skilled workers to implement.
My father realized this in the late 60s or early 70s as an engineering manager at a tech company (back when tech was Tech and tech companies actually made stuff). His solution was as follows: Line jobs were ranked A, B, and C, for high, medium, and low skill. Especially the C and B jobs, you could teach multiple jobs to a single worker, and give them a workbench at which they could do several tasks in the manufacturing process, not just one -- and have several workers at workbenches doing all these tasks. The assembly line gets slowed down when one worker gets sick, but it still proceeds because whatever job that worker was doing, their colleagues also know how to do. My father's idea was dismissed by the higher-ups to implement companywide, but it's part of the "Toyota method" now, so I guess it was independently discovered in Japan. Good ideas are like that, plus, the Japanese actually respect craftsmanship; at the end of the day, American businesses only respect money and metrics.
Secondly, how many times must it be repeated? Software engineering is not like manufacturing. It takes zero skill and next to zero cost to manufacture an additional copy of a program because a program is just bits that are trivially copied by computer. To develop new software requires a moderate to high level of skill, so you want to cultivate craftsmanship in your software developers, unless you like the current state of enterprise webshit and think that software that's slow to develop, slow to run, fraught with bugs and integration issues is just fine.
Of course you still often need those experts to have decent soft skills because many commercial projects will still need multiple developers even if everyone involved is very good. But the communication and management overheads scale far more than linearly with the size of the team so a relatively small team of very skilled developers can easily outperform a much larger team of mediocre developers in practice.
I think it does. Of course the minimal skill required is higher than hiring some person off the street to bolt on a few parts as the product proceeds down the assembly line.
I worked at Microsoft many years ago and it was a lot like a software factory
It's possible the companies who want team players are acknowledging that their programmers aren't the best so everyone needs to work together to make something that works.