I’m sure you’ve seen numerous headlines similar to this: “XYZ offloads critical thinking, other hard work to AI,” or, “I Built an AI Fill-in-the-Blank.” Of course these end up spilling into daily conversations, with strangers, or remain in the back of your mind. Especially at work, this is a topic that never ceases to be important food for thought.
AI has integrated into my work-life to the degree that I use Cursor every day, I default to tools like Google’s Nano Banana AI to compensate for skills I don’t have, or use AI tools to transcribe meetings. No doubt, there has been astounding benefits that makes me think that I can’t go back. As I’m writing, I’m sure you’re waiting for a “but”, but before that, I want to share what improvements AI has given me that prevents me from reverting to a pre-AI time.
Rubber Duck
One of the ways I first learnt to code was with the famous Harvard’s CS50 course and one of the first things Professor David Malan shared was the importance of rubber ducking. A year into my career, I discovered that talking to myself was an extremely helpful reasoning process, and it got even better when I had a friend or colleague I could bounce ideas off of. But sometimes both options become stale or unavailable.
My coworker Meghan Harris once shared that she uses Cursor to strategize her approach to a problem and it occurred to me that I could use a LLM as that “friend” to bounce ideas off of. From then on, I would try to come up with my own solution, then share that with the LLM and have it share 3 more strategies. I would pick a few that I like, then start to discuss trade-offs with it before I make my final decision of what strategy to go for. For sure, LLM has not only expanded my options, but gave me a 24/7 buddy that helps me evolve my reasoning.
Learning Tool
In a way, I view the LLM as a better Googler. Even before chatGPT, developers were already scouring the internet for potential solutions or to look up language syntax. Now LLMs do all of that faster and tailor it to your specific needs.
As a consultant, I am switching between languages, frameworks, and FullStack development. This context switching requires me to have a deeper understanding of what’s happening under the hood or quickly learn a new language/framework. To have the LLM quickly help me understand syntax or rubber duck strategies, helps free up my time to focus on the problem, rather than finagling on language specifics. Actually with this free time, I’m able to build a deeper understanding of the language/framework.
Now for the “But”
Just like those headliners, all the concerns everyone points out should not be ignored. In fact, my daily usage of LLMs makes me hyper aware. What I’ve come to realize is that any “convenient” tool comes with a greater responsibility to be more intentional with our actions. Or else, we truly might become those Axiom passengers in the movie Wall-E. Here are some pitfalls I’ve fallen into that transformed my actions.
The Speedy Solutionizer
In crunch time right before a demo or when I needed to solve a bug fix within a tight deadline, I would default to the LLM to help solve my problem. Now, I’m sure people are aware that this is bad, but during that high-pressure time, your mind does not want to think with the same diligence and tenacity it normally has, but instead might want to default to external help. Each time I defaulted, the LLM failed to provide me a viable solution. At this rate, it would’ve been faster for me to discern the problem and solve it myself. That’s exactly what I became hyper aware of.
I can use LLMs for all the benefits I mentioned above, but when it comes to solving the actual problem, I remind myself that this process should be up to me. So, to fight a natural inclination to lean on the LLM’s help, I will sometimes return to VS Code when I need quiet time away from auto-complete and resist the temptation to use the chat.
The Code Generator
They say that a software developer’s new job will be to review generated code. But how can you understand the consequence of that code, if you have not used it before? Previously, I use to explain that AI will be like Jarvis helping Tony Stark build Iron Man. However, someone pointed out that Tony built Iron Man in cave with a box of scraps.
There was one time I relied on Cursor’s generated code and it was only when my coworker Joe Chrysler pointed out a potential bug, did I realize that I wasn’t thorough enough in my code review because I did not write the code myself. Now, I believe this topic is still a greenfield and that it’s up to your discretion whether you want to still generate code and become a meticulous code reviewer or write the code yourself. For me who is not 10 years into my career yet, I do not want to default to this LLM ability just yet.
So, what’s my relationship?
Nothing is fixed yet, but what I know is that LLMs are still a benefit to my growth so I won’t be cutting it out of my work-life. But I am passionate about this job because I do enjoy problem solving. So I’m not ready for the LLM to take the joy of my job just yet.