Oh, I'm completely happy to roll all of "tooling, libraries, community" into "language". It matters a ton which one you choose!
But we're not disagreeing here. The relative merits of all these factors are expensive as hell to compare.
I'd love to have a longer discussion with you sometime around abstractions and algorithms, though. I think I have a few points on the side of abstractions that pass far beyond mere maintainability, but I also really empathize with your argument. So, it'd be interesting to flesh it all out.
I'm not sure that is well-established at all. I mean, languages probably do matter, but just how much we don't know. Also, the choice of language may well matter more due to availability of tooling (debuggers, profilers, IDEs, ease of deployment etc.) and other extra-linguistic features (GC, separate compilation, dynamic linking) than actual language features. Even measuring outputs is very hard. The average lifetime of a codebase is roughly 10 years, so the development costs must be totalled over that entire period. The cost structure however, is very different in the first year and in the eighth.
I don't think we have good answers to any of the questions: how much languages matter, how they matter and why. It is precisely because we don't know that we can argue so much :)
Some of the PL discussions do, however, make me sad because while I believe the choice of a language matters (though I'm not sure that by a whole lot, and when it does it's mostly due to extra-linguistic concerns), the choice of algorithms matters so much more. I wouldn't want young developers to think that the abstractions they use to express their algorithms is as important as the algorithms themselves (as long as the code is reasonably maintainable). Also, I don't want young developers to equate software with the code the program is written in. A useful, efficient running program is much more than its code, and sculpting code is not the sole means of achieving high-quality software.