How My Work Style Have Changed with ChatGPT

It’s been over half a year since I started using generative AI like ChatGPT for work. I still use it almost every day, to the point where I feel I can’t work without it. As a result, my work style has changed significantly. This time, for the purpose of keeping my own records, I would like to write down what has changed.

I Can Write Simple Python Code Now

Originally, I worked as a software engineer, developing business systems for about seven years. However, for the 10 years since then, I hadn’t written any code at all — I hadn’t even written a single line of Python.

What got me started was, I believe, the release of GPT-4. On a whim, I had it generate a simple data processing script, and I was moved when it worked without a hitch. Encouraged by that experience, I started trying various things, and as a result, whenever there was something that seemed solvable by programming, I naturally didn’t do it myself, nor did I write the code (I couldn’t write it in the first place). Instead, I unhesitatingly got ChatGPT or Copilot to generate the code for me to achieve my goals.

But it doesn’t work 100% as is, so while I tweak it a bit or investigate the cause of errors, I have somehow become able to read it, and I feel like I can write simple things somewhat. Of course, most of the tweaking or error investigation is also done by asking ChatGPT, and all I do is copy and paste.

Of course, what I’m doing with ChatGPT is just a bit of programming. It is by no means software engineering that ensures reliability, availability, or security, etc. However, just being able to write somehow has been a big catalyst in changing the way I work, I think.

How ChatGPT Redefined Product Management and Product Development

I have a tendency to get carried away, so when I started feeling more comfortable with Python, driven by sheer curiosity, I expanded my work scope beyond what was asked of me.

One such initiative was experimenting with AI models on my own. I had tried out models with UI on Hugging Face, but I found that just using the UI was quite limiting if you wanted to efficiently test various patterns. Particularly with Large Language Models (LLMs) and other generative AI, where new models are released almost daily, sticking to UI-based trials wouldn’t cut it for a thorough comparison and efficient workflow.

As I became more adept at writing Python code, even if it was just rudimentary, I began to integrate these new skills into my work. For instance, I would create test data with ChatGPT based on potential use cases for a product, run LLMs with scripts written by ChatGPT, and compare the outputs of different models. Without ChatGPT, I probably wouldn’t have gone beyond a cursory trial based on what I read online or a few random tests — barely scratching the surface of understanding.

Once I got the hang of scripting (or generating scripts), I started to create prototypes for new products and features on my own. Especially when planning products involving generative AI, it became clear that extensive initial planning documents were somewhat futile. Instead, assessing the quality of the output and the user experience it facilitated was paramount.

With tools like ChatGPT and Copilot at my disposal, I could handle the initial prototyping phase solo. This approach meant that by the time we moved to development, many functional requirements were already defined to an implementable level since I was working with operational prototypes. Of course, prototypes rarely transition into real services as-is — there are non-functional requirements to consider, and the actual services may not even use Python. Yet, prototyping is vastly more efficient and effective than planning with documents alone.

Certainly, there’s a plethora of ideas that end up getting discarded. But since I’m doing this solo, if something doesn’t quite hit the mark, I can drop it without a second thought, devoid of any emotional attachment.

Moreover, I’ve taken over most of the post-release data analysis. From writing SQL queries to scripting the combination of data from multiple sources for analysis, including parsing unstructured data like app reviews, I handle it all. Data analysis is often an iterative process of hypothesis and discovery, and since it’s just me, I can dive as deep as I wish.

Starting with hands-on trials of AI models for direct understanding, to crafting prototypes grounded in technical and user insights, followed by analysis — I rely heavily on ChatGPT and Copilot throughout. In a larger organization, these tasks might have been shared among many, and I might not have relied on ChatGPT as much. But in a startup environment, resources are scarce, and leveraging ChatGPT has become a necessity — a necessity I try to view as an advantage.

My Scope of Work Has Expanded, Workload Increased, but Working Hours and Frustrations Have Decreased

As I mentioned earlier, the scope of my activities has expanded. However, I’m not engaging in the deep development work typical of a software engineer, nor can I precisely categorize what I’m doing now as a Product Manager’s job. I’m not entirely sure if the role I’ve taken on has a proper title or if the way I’ve expanded it is correct.

But, I do believe that, as a result, my productivity has significantly increased — though this is just a guess. And personally, it’s quite fun to handle a variety of tasks.

Naturally, as the range of my responsibilities has grown, so has the volume of my individual workload. On the other hand, I feel as though my total working hours have decreased. That’s how much I rely on ChatGPT/Copilot. Thanks to these tools, I’ve honed my skills in extracting desired results from even the most rough prompts — my ChatGPT history would probably look like the exact opposite of the prompt techniques commonly discussed online. (Of course, I ensure to craft them properly when integrating into a product.)

Also, as expected, being able to handle more on my own has greatly reduced both communication costs and my own sense of uncertainty or frustration. By ‘frustration,’ I don’t mean the kind stemming from interpersonal relations but rather the self-imposed kind — the feeling of ‘I wish I could do this by myself’ has significantly lessened because now I can actually do it myself, with the help of ChatGPT.

Take the earlier example of prototype development: from a PM’s perspective, I have many ideas I want to try but can’t create a working prototype, and I can’t articulate everything clearly yet. From an engineer’s point of view, it’s not very interesting to be roped into someone else’s trial and error, and while there’s not much technical intrigue in prototyping, it’s part of the job (I remember feeling this way myself when I was an engineer).

Being able to do this alone as a PM seems to have reduced not only the total amount of communication but also the communication that is prone to cause frustration. (Though, everyone is very nice, so they may not feel like me.)

Got One More Job

As I have been doing such tasks, I started a side job for the first time in my life. I am supporting the promotion of how to utilize generative AIs like ChatGPT in business operations and products. Perhaps I was recognized for my experience in utilizing generative AI in products even before ChatGPT was released, but the know-how I’ve gained using ChatGPT is directly applicable in my side job as well.

rinna is still a startup and a small organization, but the place where I am doing my side job is a much larger entity. At rinna, I am in a management role, while at my side job, I am an IC (Individual Contributor) where personal contribution is expected. While the necessary knowledge for utilizing generative AI is very similar, the culture, expectations, and roles are completely different. Working in different roles helps to keep any complacency in check and to discipline myself, which I find very meaningful. However, my workload has just been increasing…

At the End

I do not think that using generative AI is incredible or that not using it means falling behind. However, for someone like me who has many things they want to do but isn’t particularly fast at doing them, ChatGPT and Copilot, which can handle many tasks in almost no time, have become remarkably dependable tools, to the extent that it’s hard to find the right words to describe their utility.

In summary, what has changed for me in the past over six months since starting to use ChatGPT is that I became a Python beginner, my work content and workload have increased, and as a result, my intellectual curiosity has been satisfied and I have created an environment that encourages self-discipline.

Considering the changes that have occurred in this past year from the year before, I have no clear idea what the next year will look like, but I intend to make the things that can be done easily as easy as possible and to devote my time to enjoying the many changes that will happen in the future.