Small<br>LLMs have brought about enormous change in the software industry, despite contrarian claims by the modern code-hipster that it hasn't. Now, that isn't to say that the studies (1) claiming LLM usage is not a productivity gain aren't true or are poorly formed or anything to that effect; rather, it's that the vast majority of developers and LLM users in those studies are the ones at risk of being replaced because of the nature of both LLMs and software engineering as a whole.
It's worth rewinding nearly a decade and a half to the start of my career as a software engineer. At the time, in a fairly well known liberal arts college, the majority of students, especially the elite math, econ, and physics majors, went into consulting. Computer Science as a subject was on the cusp of exploding in popularity, however was still among the smallest "scientific" majors. I believe my graduating class had about 7 majors, whereas the Physics department, the next largest, had close to 30.
This is a time when software wasn't cool, The Internship hand't been released yet, and, frankly, people didn't really get how nice the lifestyle of a software engineer can be. Generally, people who were in software were people who genuinely loved solving problems, building things, and working with computers. People who wanted to make money still went into finance and consulting. Software at this time still hearkened back to a time of hacks and optimizations to squeeze every drop out of the rare machine resources we had available.
What changed, to me, is unclear. Perhaps Silicon Valley, perhaps the word just "got out" about the pay and lifestyle, but the mid 2010's saw a swing. "Code Bootcamps" were born, software ate everything, and companies wanted anyone who could type on a keyboard, and Big Co. would pay six figures to get them. I have at least a few friends who went into software simply because they heard they could make a lot more money. While "a job is a job" and "work to live, don't live to work" are both laudable mantras, they do not mesh with sitting at a computer for eight hours a day, working through hard problems, and dealing with stubborn frameworks -- that's where we start seeing burnout.
Back to the present, or recent past at the earliest, these same engineers from the mid aughts are now "seniors." Those who clocked in on the 9-5 and didn't care to learn the newest technology, read a blog about how postgres deals with indices, or build an open source project are now the ones leading and designing projects, reviewing code, and making large scale technical decisions. These are the people that are using LLMs and not seeing the benefits.
It's not that LLMs are not useful to everyone, it's that they are far more powerful to those that have honed their craft and can, more-or-less, visualize the exact structure that they want. The LLM is not a tool for ideation, as much as it is a powerful tool for executing vision. Engineers who built great software in the past will continue to build great software, just an order of magnitude faster.