The fear of AI fatigue proved to be unfounded

01A farewell to Stack Overflow#

When ChatGPT was introduced in late 2022, I remember that the biggest shift for me was the daily trips to Stack Overflow being replaced by the minimal black-and-white chat interface.

I don’t know if I miss the feeling of the endless hours being stuck, searching on Stack Overflow for a solution to my countless bugs. But I sure do miss the feeling I got when I finally found the answer, buried deep beneath the surface, and saw that missing piece of code make my project run again.

02The AI fatigue#

It felt like the adoption of ChatGPT in our daily work happened overnight. Not only was ChatGPT replacing Stack Overflow for developers, but the whole organization was also handed a tool that provided brainstorming, planning, rewriting, proofreading, and much more.

Not long after, the number of articles, blogs, and videos with the shared theme of “If you are not doing this…” exploded, which turned into a daily battle of trying to stay on top of the newest trend within harnessing, loops, agentic workflows, context, prompt engineering, and so on.

While we spent more time and more tokens on more agents, I felt like we weren’t getting the outcome to justify the effort. As we were chasing the dream of a fully automated agentic workflow, the endless stream of AI FOMO articles kept coming, making it feel like the amount of value you could produce was only a matter of how many tokens you could afford. And we were the only ones not “harnessing” the full power of AI.

We started to fear that it was only a matter of time before we would be left behind by another agency that was able to harness the magic of agentic workflows. But that fear turned out to feel somewhat unfounded.

03The AI maturity#

Even though the AI FOMO articles were still flooding the internet, and the race for bigger numbers in benchmark tests continued between Anthropic, OpenAI, and Google, we were starting to settle into the new era of AI coding.

We found that the principle of simplicity was still in full effect, and doing less was often the answer for us. The goal was not to have AI running on autopilot, but rather for it to be an efficient tool for developers.

We settled on a few skills and a few MCPs that created a workflow allowing the AI to be productive within the boundaries we defined. And even if we weren’t reading through all the code the AI produced, we still had a good understanding of how the different bricks and pieces were assembled.

But at this point, I also started to miss the era before the emergence of AI.

04The joy of coding#

If I had thought about the capabilities of AI before it became part of my daily work to the extent that it is today, I would probably have loved the idea of having an agent doing all the heavy lifting, so I could focus on the bigger pieces related to architecture, infrastructure, and the functionality of the features I was building.

But I really do miss the long coding sessions, grinding through line by line, constantly optimizing, renaming, and refactoring in pursuit of the unreachable perfect state.

The best way to describe that feeling, and the best way to end this post, is with a quote from Zixian Chen. Until next time.

The way I keep describing it in my head is that it’s like my child disappeared and got replaced by a gorgeous kid I don’t recognize, where I didn’t go through the growing pains with this one, and even though it’s objectively better, it doesn’t really feel like mine.

https://zixianchen.com/blog/feeling-some-ai-coding-fomo