I think that's the point the author made. If the big majority of users wants this, but software developers want that, they obviously focus on this. Its what recent history confirmed and its what's logic in a capitalistic standpoint.
To break it down, developers want intelligence and quality, users want patience and validation. ChatGPT is good at the latter and okay (in comparison to competitors) at the first.
I do agree that ChatGPT may just be good enough for casual users to not be worth exploring (I'm tired with constant churn of AI releases too - on that note, there should be a worldwide ban on multiple AI companies releasing similar tools at the same time; I don't have time to look into all of them at once!) - but they're definitely not getting the suboptimal deal here. At least not ones on the paid plan that are aware of the model switcher in the UI.
EDIT: Also, setting gpt-4o as the default model gives ChatGPT another stickiness point: it's (AFAIK still) unique image generator that qualitatively outclasses anything that came before.
I want to challenge this assumption. I think ChatGPT is good enough for the use cases of most of its users. However, for specialist/power user work (e.g. coding, enterprise AI, foundation models for AI tools) there is strong pressure to identify the models with the best performance/cost/latency/procurement characteristics.
I think most "vibe coding" enthusiasts are keenly aware of the difference between Claude 3.7/Gemini Pro 2.5 and GPT-4.1. Likewise, people developing AI chatbots quickly become aware of the latency difference between e.g. OpenAI's and Claude (via Bedrock)'s batch APIs.
This is similar to how most non-professionals can get away with Paint.NET, while professional photo/graphic design people struggle to jump from Photoshop to anything else.