State of AI 2025

This thing opened up with something that I couldn't relate to: putting questions, big or small, into a generative text tool to get an answer from. I understand that Google, Microsoft and Apple are educating as many people as possible that this is an okay thing but we don't have to keep parroting it.

Anyway.

I recommend walking through the results: the demographics, the disabilities markers and the like. I was a bit shocked to see Neovim on there but it is a pretty flexible editor. As expected and dictated by market dynamics, Microsoft's tooling — through GitHub CoPilot, ChatGPT and the like, all dominate in usage and deployment. When you can make it free and use said free input to refine their work, it begins to pay for itself — according to economist/trickle-down "theory" logic. There's also a bit of a bias in that most people only have used one coding "agent" and you can't filter against that^[1].

Something I expected was that it'd be predominately the language that's the most difficult to document that'd be at the top of this list. I believe that languages that have baked in documentation have less of a need of discovery for tools like it. That and the difficulty of building for the Web puts that language higher on the list of "how the fuck do I do this" (the whole point of generative code tool). Accessibility to the runtime is another factor, perhaps, but negligible. More folks need to use TypeDoc; my $0.10 (ignoring tariffs and stagflation).

Half of the use of refactoring use is either because folks can't read the code or they want to rename variables (which folds into the first, IMO). That explains why the happiness marker seems way more optimistic than it should be. The note on "strong consensus" of AI integration into workflows is notable as well as folks mentioning increasing productivity. Does productivity means more time spent "doing things" or less time getting to something "productive"? Additionally, removing the neutral option to be a 4-opinion split regarding AI existential risk might have leaned it towards more of a "no" just on the fact that developers tend to underestimate things. The reaching AGI part is just embarrassing, tbh.

This ending with a AI-promoted-in-young-classrooms note told me what I needed to know about this survey: a soft means of reaffirming Microsoft's tool integration into the lives of developers (since we collectively refuse to leave the ecosystem) by means of labor automation.

Thank you, next.