I was talking with a friend about development when they asked me a question.
“What was the hardest problem you faced, and how did you solve it?”
The first thing that came to mind was the work I did to improve music search. I could probably explain what I built and how it worked in the end. But when I tried to explain exactly what was difficult, what I thought the cause was, and why I chose that approach, I got stuck.
I had clearly done a lot of work and built things, but when I tried to put that time into my own words, it felt harder than I expected. I wish I could take just one problem I had worked on and explain it from beginning to end: where I got stuck, what I checked, and why I reached my conclusion.
Finishing a task and understanding it are different things
At work, I move between several tasks. Understanding one thing deeply takes time, while new work keeps coming in. At times, I got used to asking AI, implementing the answer, and moving on as soon as it worked. The things I meant to revisit and study later kept getting pushed behind whatever came next.
I still want to keep using AI’s help. It gives me a starting point when I do not know where to begin, helps me read unfamiliar code, and helps narrow down the causes of errors. It also lets me quickly turn an idea into something real and see whether it is actually useful.
But checking that the code runs is different from checking that I understand why it works. An explanation can make sense while I am reading it, yet leave gaps when I put it away and try to explain the whole thing myself. I want to look back not only at the fact that I used AI, but also at how much of the result I checked and what I thought through along the way.
I was reminded of the time I built my blog. I enjoyed looking up what I needed, applying what I learned right away, and making things. I wanted to organize the code well and understand how what I had built worked. That is the kind of time I want to bring back now.
That does not mean I plan to write every line by myself or understand everything related before moving on. Even a small feature can raise endless questions. If I set a rule that I have to understand everything perfectly, I will probably feel overwhelmed before I start. I want to begin by choosing one problem I wish I could explain.
One problem a week
From now on, I want to choose one problem I worked on that week and write it down. It can be something I solved but cannot explain well, or something where I chose one of several approaches and want to understand the reason for that choice better.
The topic does not need to be grand. “Understanding a search system” is too broad to know where to begin. But if I narrow it to a question like “Why does this result appear first when I enter the same search term?”, it becomes a little easier to decide what to check first.
After choosing a problem, I want to write down what I already know before asking AI to explain it. I can note the order in which I think the code runs, why I think the result appeared, and what might change if I used another approach.
The writing does not need to be polished. I want to leave sentences like “I understand up to here, but I do not know what happens next” or “I think this part is responsible, but I have not checked it yet.” Writing down my first thoughts will let me tell later what I knew already and what I learned from the documentation or AI’s explanation.
Then I want to fill in the gaps by looking at the actual code, the relevant documentation, and AI’s explanation. If the answer still does not make sense, I will narrow the question further: “What would happen if I removed this part?” or “In what situation would this cause a problem?”
Check it on a small scale
When an explanation alone is not enough to feel certain, I want to check it with a small experiment. I can isolate the part I am curious about in a testable environment and see how the result changes when I change one condition.
If something runs twice when I expected it to run once, I can first check when it runs and then change one part I believe is the cause. I can compare the results before and after the change to see whether my thinking was right.
Before an experiment, I also want to write down my prediction in one line. If I run it without thinking about what I expect, it would be easy to come up with a plausible explanation after seeing the result. If the result differs from my prediction, I want to look back before searching for the next answer: Was my first hypothesis wrong? Did the changed condition affect something else? Did I miss something in the way I checked it?
Of course, I cannot run an experiment for every problem. In those cases, I want to separate what I confirmed in the documentation from what I am guessing. Even being clear about what I know so far and what I still need to check gives me somewhere to look next.
What I want to leave on the blog
I want to start each post by explaining what the problem was and why it was difficult. Then I want to write down my first hypothesis, the approach I chose and why, and what I confirmed myself. I also want to leave the parts I still cannot explain well and the questions I want to explore next.
If I only summarize the solution, it is easy to lose the thinking that happened along the way. If I tried another approach, I want to include why I changed it. If the result differed from what I expected, I want to write down what I learned. Not every post needs to end with a neat conclusion. I can write what I confirmed that week and carry the remaining questions forward.
After writing a post, I want to return to the question I started with.
“What was the hardest part of this problem, and how did I solve it?”
I want to check whether I can explain the rough flow without looking at the post. If I get stuck again, I can learn more about that part.
This is a plan I am only starting now. Once I try it, I may make the scope of a problem too broad, or it may take longer to write things down than I expect. I will adjust those parts as I go.
For now, I want to choose one problem and start by writing my own explanation. I want to gradually expand what I understand inside the code I built with AI’s help. When I read it again later, I hope it will show what I was thinking about and what I learned at the time.
