Is anyone actually using the OpenAI for production apps?

ChatGPT is obviously very capable but given it's an massive general-purpose LLM, the current version tends to be slow and expensive for narrow tasks. Even a product like Github Copilot is actively losing money as it spends more in API fees than it gets from subscribers.

Narrow use-case like parsing a PDF, for example, does not need a model with all humanity's knowledge to be effective. I know the Assistants API is meant to give the GPT models a more narrow focus but all this does is introduce persistent context to the model; the underlying workhorse is still the massive LLM.

There's also the issue where, if you're building a production-grade solutions that you intent to launch and/or sell to businesses, you run into some significant issues in terms of data protection and privacy since all your clients' data will ultimately flow through OpenAI. Ignoring the massive ethical and social implications, this may simply be incompatible with your clients data protection policies leaving you with a product that's impossible to get past procurement.

And lastly, using the OpenAI API to build on top of the GPT models effectively just boils down to creating wrappers for the API. There's barely any innovation or development to be done in the first place, leaving you with a product or solution that anyone with even a basic skill set to copy and reproduce. This is perhaps the most damning aspect since without the potential to actually create anything new or useful, there's really no point making anything at all. The only value you're creating is porting ChatGPT to users' applications directly, instead of them having to go to the OpenAI website. That's worth something, but it's a much less sophisticated project and once all the integrations are made for the most popular platform, there's simply no other work to be done.

Given all of the above, is anyone even bothering to build on top of OpenAI, and if so, what are some projects you're working on?