The Developing Developer

Recently at my workplace we have started developing using Cursor, a fork of VSCode that integrates Anthropic’s Claude-3.5 model, and after about a month or two of developing using this tool I have some initial thoughts, fears, and predictions for the role of the average Developer. First, I believe that the capabilities of this model make programming significantly more enjoyable. Second, I fear for not the Developer career-wise, but more capability-wise, and finally, I believe that using Cursor has reframed my idea of not only what a Developer is, but what a Developer will be if coding assistants continue to improve.

High Praises

First things first, obligatory props need to be given. the rate of growth for some of these LLM models are incredible, AI has become an insanely useful tool for learning new things, creating boilerplate, and unit test generation. However it goes further than that, with new AI-integrated IDE’s such as Cursor and Codium, coding with an AI companion has never been easier, but what does that mean for the modern Developer? Countless demos of cutting edge AI creating apps with only a few prompts have surfaced recently, and the chorus of individuals warning of the AI end times have never been louder.

Some say that it’s “the end of the computer science industry”, but I’ve come to a different conclusion. Sure Anthropic’s Claude-3.5 has performed extremely well on coding questions, and I will not attempt to downplay Cursor’s capabilities. In fact, on one occasion I was looking at integrating a plugin for a Leafletjs map, it was last updated almost a decade ago, and was written in jQuery. Not a quick adaptation to Next.js, or so I thought. jQuery is a javascript library, and Next.js (while configured for typescript) was not a long shot from being compatible. So in the interest of saving myself some time, I thought to give Cursor an opportunity to wow me, and was not disappointed. Cursor was able to convert the entire script (~250 lines) into a Next.js component that worked seamlessly with Leafletjs. However, testimony aside, AI will not be the downfall of the Computer Science industry. In fact, AI will actually be a catalyst that pushes the role of the software engineer to new heights.

The Evolution of the Software Engineer

Technology is an incredibly volatile field. The standard Software Engineer has heard of if not already used at least 3 different Javascript Frameworks. Frameworks that accomplish the same task either marginally better at best, or harder at worst. Our landscape is constantly changing and AI is a tool that will continue to squeeze its way deeper and deeper into our workflows. Yet I don’t fear it replacing Software Engineers, in fact it may help to reduce time implementing simple tasks such that Engineers can take the time to plan larger, more complex software. One of the most important parts of any large project is the thought behind the solution. Many times I have taken a step back before integrating a big feature, and thought to myself, “just how am I going to do this?”, it’s that thought process that makes a Software Engineer. Right now the reason that LLM’s are doing well is because programming itself is a language just like English, Spanish, or any other. Therefore writing the language is not the hard part of the equation, it’s the grouping of sentences and words to form a succinct and coherent point. In the Software Engineer’s case, it is the ability to accomplish a goal bearing in mind space complexity as well as speed (and that’s a simplification). It is not the ability to write that is the software engineer’s greatest ability, but the organization/problem solving mindset.

Moving in the opposite direction of progress, and going back to more ancient times, the written word was something wonderful to behold. The ability to form a thought and record it on a piece of stone or parchment for others to see was truly a feat. There were even scribes whose entire job were to read and write. Yet as time progressed, the written word became much more prevalent and accessible, but notice we still have authors, writers, and poets centuries later. Originally, it was the ability to write that was important, however as time progressed it was the ability to write well that stood the test of time.

The Pitfall to Avoid

That being said, I fear Programmers will start relying too heavily on these technologies. As of current day, coding assistants are great for finding syntax errors and writing ~10 line chunks of code; with heavy instruction of course. However, it is upon relying on these tools for whole features and complex issues that severe weaknesses start to show. Mainly in areas like multi-threading, complex codebases, and anything that isn’t in Javascript/Typescript or Python (where I suspect most training data comes from). I’ve tried to dive deep and push the limits of what these tools can accomplish, and I’ve noticed from myself the pitfall of just throwing Hail Mary queries to Chat-GPT or Claude to avoid doing any real work. Whenever I start doing this the quality of code in the my projects immediately begins to nosedive, and I have to take a step back and remind myself that I have a brain and should use it. I’ve since pulled myself out of the pit but I fear for Developers falling into to the same trap I have. As previously said, the role of the Developer is to determine the best route to accomplish a goal in the quickest and most proficient way possible. Giving up that role and letting AI hallucinate the correct solution is a fast pass to bad code, confusing codebases, and not getting the job done. There was one instance where a classmate and I wanted to get a school project done with only one more function to implement, feeling burnt out we tried for almost two hours to help AI finish the job, after trying so long to avoid thinking, we figured we had to do it ourselves.

We had a working solution in under ten minutes.

Banking on AI not only got us further away from getting the project done, but if we took the time to actually think about what we needed to do we could have saved hours. Now imagine if instead of two students trying to avoid school work, it’s DoorDash, Microsoft, or Trading Firms. The quality and speed of development will drop in spectacular fashion, which is why I fear for the capabilities of Developers but not for the career path itself. If anything AI assistants will help the cream rise to the top even more.

Final Thoughts

AI assistants are here to stay and are only going to become more embedded into our respective workflows. However I do not fear for the industry’s stability in the least. Developers will still be needed long after AI stops trying to rewrite the whole file to fix one line. I have praise for how it can help Developers, faith that the software engineer with evolve alongside new technologies, and a twinge of unease for more reliant Developers falling into the illusion of AI’s power. But overall I believe that just like the scribe, Software Engineers will continue to be the poets of the code, and no amount of pattern recognition can change that.