Why AI coding tools are output multipliers, not substitutes for human direction, innovation, and taste.
The first programmers to adopt COBOL, Fortran, and similar high-level languages were surely looked down upon by their assembler-swinging colleagues. I can hear the reverberations, echoing from the 1950s, of the presumed end of the art of programming in light of these new higher-level languages. Obviously, pulling up the abstraction level did not end programming as we know it; rather, it started it more than anything.
I see the same momentum with AI coding agents, which in just a few years have become the de facto code writers and reviewers of modern software production. Looking closely at our engineering team of about 100 developers, I can see that a third are bullish adopters of each new incredible wave of innovation.
Many in this group are senior, experienced engineers who just found a way to multiply their impact on the company many times over. It is amazing to see how productive and innovative senior and experienced developers are in combination with the new tools. It's almost like watching them develop superpowers.
We also see a group on our team that is not yet fully on board with using AI coding tools to their full extent. There are concerns ranging from code quality and maintainability to understanding the ramifications of a codebase that has been optimized and expanded over many years. For some, it doesn't feel right to let an AI coding tool loose on our most prized assets.