This is an insightful read, but what makes it interesting for me is the question around what abstractions do for us.
Here are some thoughts from me.
This post inspired me to spend a bit of time this morning learning about Pydantic and FastAPI. I know a lot of python, but I’ve not been keeping up due to the nature of my role. I’m sure these libraries are well known to modern python developers, but I’d not spent any time at all. I used OpenAI’s new study mode this morning to get the gist of Pedantic. It abstracts away the complexity of type checking your data models, and hooks directly into FastAPI to provide very nice error messages if you send incorrect or missing data into one of the APIs that you build with FastAPI. This is a kind of nice layer of abstraction, moving you away from the code to what you want to implement.
This is a theme that is mentioned a bit more in the post that I’m linking to. What was so fascinating is that by leveraging on top of some common abstractions OpenAI could move at pace in an almost uncoordinated distributed fashion, and behind all of that the models are doing a lot of the work.
Really fascinating is the discussion around their core back end infrastructure. They use just a small number of core Azure primitives, and with effectively just a container environment, a blob storage system, and a database, they have built this huge ecosystem. It really is the case that they are building wrappers around GPUs.
The other thing that was amazing to read about was how a small team delivered a milestone feature, in just seven weeks.
One thing they don’t have is the a challenge of the number of services they need to support being far greater than the staffing levels that they have to support those services. They are in a growth mode, so they have the luxury of being able to throw some of the best engineers on the planet at problems, and do so in parallel. Most of the time I think about code a liability, but in this environment code is possibility. It must be simultaneously be an exciting, frustrating, scary, and intense place to work at.
found on 2025–08–02