The advent of AI coding agents has completely blown up all p…

Well it has made them much less important. They are no longer a medium to think with (if they ever were), more like a construction material.

And you, the developer, are no longer hands-on with these languages, suddenly your role (if you still have one) is more managerial. You tell the workers (agents) what to do with the code, and maybe you'll look at the final product but mostly not. Everybody got promotions, and Peter Principle applies, so most developers aren't good managers and suddenly find themselves at their level of incompetence.

Everybody who wants to keep working is forced to create a new methodology of their own – different tools, methods, standards, practices.

• or adopt one, I suppose most people still work in big companies, where the elite levels decide on a methodology and instruct their underlings to use it.
• we must mention the AImish who just refuse to adopt the new methods. So, no, not everybody.

  • Rich Hickey might be the most brilliant mind in software today, and he is a stark refusenik 'AI' is Not a Tool by Rich Hickey · GitHub Clojure - Contributing

    • No Generated Code: Clojure’s code is written and reviewed by humans. Code generated by a large language model or similar technology, such as Anthropic’s Claude, GitHub/Microsoft’s Copilot, OpenAI’s ChatGPT, Facebook/Meta’s Code Llama et al, is not compliant with the covenants and representations of Clojure’s Contributor’s Agreement, and is thus not acceptable as code for Clojure.

But we still use the same languages. Whether because of community size (Python) or conceptual integrity (Clojure) or performance or expressivity or whatnot, LLM coding is still basically a process of producing artifacts in the same old languages, just with a different process.

I suppose agent-based architectures are the new languages, there are lots of these but for the most part I think are not that important because of The Other Bitter Lesson. They strike me as ways to do traditional computational system building, where you have modules and API conventions that connect them.

Programming Languages used to be at the forefront of centaurification. That is, they were the medium through which human and mechanical intelligence collaborated with each other, how ideas became concrete computational realities.

But that's no longer the case. They still are important in that process but they've been effectively demoted. Now we use natural language specs and prompts to create software, agents write the code, maybe we look at, maybe not.

And the choice of language matters much less than it used to. I mean, the ecosystem considerations are still there, but as for esthetics and ways of thinking, who cares? A generated python library is just as good (better pragmatically) than a clojure library even if the latter is more elegant. Elegance doesn't matter so much (if it ever did – it meant a lot to be personally but didn't seem to effect general practice much.)

It's a well-known truism that software modularity boundaries mirror organization boundaries Conway's Law But LLMs dissolve both. Most of the current structure of corporations and of software is probably obsolete and will crumble away as soon as the new structures start to emerge.