A tweet that got 15K likes and reminded me of Fortran (created to let scientists and engineers program at a higher level) is below.
Computer science is gradually returning to the domain of physicists, mathematicians, and electrical engineers as large language models automate much of what we currently call software engineering.
The field’s center of gravity is shifting away from manual code writing and toward deeper theoretical thinking, mathematical insight, and systems-level reasoning.
This is a very interesting thing to think about. The one thing I haven’t experimented with is starting a vibe coded / AI assisted project without thinking about the architecture I want to follow. Since I’ve spent most of my career developing things for GPUs I have a very dead-set architecture I follow for data handling and thus I try to be very modular with how I built new physics.
This ends up making my application easy to modify and enhance. I wonder what kind of architecture will a LLM follow without this explicit guidance. Indeed when Fortran came out and let scientists and engineers write code they mostly wrote what we now consider to be terrible code. Will LLMs be able to bridge this gap and not only write good code but also well architectured code?
For the projects in which I’ve set the architecture, using LLMs has been a game changer because I can progress at a speed I wasn’t able before. I can leave something running while I cook dinner and come back to a tested implementation. It is a workaholics fever dream.
I think right now the “domain knowledge” still includes the design and architecture of the program, in addition to physics or other fields. But the coding itself can now be automated, which is amazing.