APIs Are Temporary
Discover the risks of relying on third-party APIs. Learn valuable lessons about API dependencies and planning for change.
A REST API has always been a great enabler for new projects. I suspect that at this moment, with LLMs being new and enabling so many ideas, that we must be at an all time high for small products being offered where the product is almost entirely a wrapper around an API. There is nothing wrong with this, and ChatGPT has become the de-facto LLM API format in the same way that S3 has for object storage. So, if you have to jump providers at any point it should be as easy as changing a URL.
I worked on a fairly successful Stock and Crypto Chat Bot for a few years that was just a wrapper. Sure, I did a lot of sleuthing of data to make it presentable. There was also a lot of time spent intelligently caching to save on API calls. However, at the end of the day all of the heavy lifting was being done by exchanges. On top of that the bot only worked on Discord and Telegram meaning that I was completely beholden to them too. I can't bring up the Discord API without mentioning how extremely hostile towards bot developers they were. It was not a fun platform to support. Despite impressive growth and a clear path to profitability, the product ultimately failed due to its heavy reliance on third-party APIs.
What could I do when Discord took away key interactions that killed usability, ask users to make an account somewhere else?
What could I do when US Stock Market data pricing sky rocketed, run my own cable to the New York Stock Exchange?
What could I do when the Cryto hype cycle crashed and no one wanted to provide data anymore, run my own nodes?
Realistically, my stock market bot only worked when those services were provided to me. I could have found work arounds or stretched the rules a little on my licensing agreements, but to what end? As long as the product couldn't stand on its own the cycle would continue.
Every project that relies on a 3rd party to do something critical that they can't provide for themselves should consider their days numbered. I would be especially scared of LLMs because I think that right now the entire market is being heavily subsidized by Venture Capitalists and Established Tech players. Unless GPT5 sees the same exponential growth (I think it will) then it isn't even remotely profitable to run your own inference hardware for 99% of use cases. LLMs are still far too dumb for the value they provide.
In conclusion, while APIs can be a powerful tool for creating new products and services, they also come with inherent risks. Developers must be aware of their reliance on 3rd party services and plan accordingly. LLMs are incredibly hot right now so it is easy to pick on them, but it is important to know that no level of power makes you safe from being shut down when your product relies on an API. Reddit and Twitter 3rd party clients are a great example of a universally loved product disappearing because the host decided it no longer needed to provide options. No one cries over Flamingo or Tweetbot anymore, and soon no one will even remember Apollo.
Good luck hacking.