The major use cases have not been filled, only the major corporate use cases. That doesn't mean there aren't still major gaps to be filled now, or that the current use cases won't evolve in the future.

This talk argues there are many purposes for formal languages other than programming: https://medium.com/bits-and-behavior/my-splash-2016-keynote-...

So it may be true that "programming language" development is done, but there are many more applications for formal languages than just writing apps and websites and such.

Moreover, it may be true that less code as a proportion of all code is being written or read by humans. But more humans today are reading and writing code than ever before. So the ability to express unambiguous ideas to agents using formal langauges is more important than ever, even if people are only reading/writing a small proportion of all code.

Similarly, it's not true that because agents will be reading / writing code that language ergonomics do not matter. Agents have resource restrictions like token budgets, and they are also trained using human knowledge and conventions. Ergonomics apply to agents just as they do humans; languages which are verbose require more tokens to write, languages that can be verified are more useful in agentic loops, languages with live compiler introspection and interpreters can give fast results to agents etc.

So I hold exactly the opposite view: it's not clear to me why languages built 30 years ago in a world where the Internet barely existed would be accidentally ideal for AI agents in 2026 to the pint we can say the whole field is solved for good (and no, I don't think the amount of training data on those languages seals the fate because that status is ephemeral and also it's not clear training on one language corpus isn't transferrable to other languages).