I've had this attitude before and missed out on some major opportunities. For example, even though I was an early smartphone adopter, I refused to develop apps for the iPhone when the AppStore was launched in 2008 because of the closed nature of Apple's ecosystem. There are a variety of billion dollar companies which can attest that building their castle in Apple's kingdom worked out fine for them.
The big question today is: Do you try to make an AI business using OpenAI's APIs, or do you host everything yourself? One could make the argument either way.
Most business advice is really good at figuring out in hindsight why things went a certain way. But usually it's not that great at predicting what will work in the future.
This is a good counterpoint. I fell for this too.
There is an argument for airbnb the lands with a castle on wheels.
You use their APIs in a way that commoditizes them. Ideally your customers don't care if you switch to Anthropic, because the LLM provider is not the reason customers are picking you. Likewise, there is some structural reason that OpenAI will never release a feature that rugpulls you, eg, no 'chat to your PDF'.
An extreme form is self-hosted on edge-only devices where folks are buying some other hw. Ex: Nvidia selling GPUs and giving out free Triton inferencing OSS software. But most are in the middle, eg, some accounting app now with LLMs. Our case of investigations in louie.ai is right at that boundary: OpenAI likes to support data analysis, but folks using Splunk/databricks/etc all day expect a lot more out of software here, and that's too at-odds with OpenAI's org chart and customerbase.