LLMs as Language Compilers: Lessons from Fortran for the Future of Coding (cyber-omelette.com)

Large Language Models (LLMs) have undergone a transformative evolution, shifting from simple chat interfaces to sophisticated coding agents capable of autonomously completing programming tasks that once required entire engineering teams. This shift is exemplified by a significant decline in Stack Overflow usage, with new posts dropping 77% since 2022, as developers increasingly leverage tools like ChatGPT and coding agents to expedite their work. One developer shared their personal success in using these agents to build a full-stack iOS app, highlighting both the productive capabilities and the limitations of current AI tools in programming. This resurgence in automatic programming echoes historical developments in computing, particularly during the 1950s with the introduction of languages like FORTRAN, which revolutionized coding efficiency by abstracting complex machine instructions. Despite initial skepticism and fears of obsolescence, history shows that simpler languages did not eliminate the need for programmers but rather expanded the field, leading to a dramatic increase in the number of computer workers. As LLMs and coding agents continue to evolve, the AI/ML community is faced with the dual challenge of harnessing these advancements while navigating the complexities that remain in programming—ushering in a new era of innovation reminiscent of past technological revolutions.