models get better, agents get smarter, rigid workflows become less useful. if your startup’s value gets destroyed by a step-function improvement in models you’re in trouble, if it improves from plugging in the latest sota model.. that’s value creation. 
lots of value left on the table for automations at scale even if seemingly small.

I think bring your own model (or choose from a set) is going to get more and more popular, especially with the rate they're changing, getting better. They also have more subtle differences than most people realize, imo.

I write code and sometimes stories. jakeacasey.com

yea problem is developing the infra to support the server call needs of each model and changing the pricing (token consumption) per and rate limits etc.. that comes a long with that. I think there's genuinely a middle ware business here.

we just average it out lol, it ends up costing us more for the more expensive models but the pricing of all these models is already so generous tbh.

I own https://nimgpt.com/ and https://dirigiblestudio.com/!

haha true, but i'm thinking from a CS pov. depends on the end -user reqs and model limitations + at what complexity the model starts to break (warning users // outright blocking functions)