Yew Jin Lim

I recently placed top 20% in a Kaggle competition. But I didn't write a single line of feature engineering. I didn't tune a single hyperparameter.

Instead, I built a system that knows how to approach ML problems - then pointed it at the competition and watched.

When something broke, I didn't fix the bug. I asked: "What was missing from the system's understanding?"

Then I taught the system that knowledge and it persists. The next competition the system starts smarter.

This is the shift I keep thinking about: we're moving from solving problems to solving problem-solving.

The code is becoming the comment. The solution is becoming the process. The programmer is becoming the teacher.

Read this in detail in my latest substack article:

I placed 4th in a Kaggle competition last month. I am not a data scientist and barely know data science.

That gap should worry you.

I built an agentic workflow that handles experiments and iterations. The AI writes the feature engineering code, tunes hyperparameters, manages GPU instances. Thousands of lines of code I never wrote.

What did I do? I noticed that missing values were the signal. I caught a leakage bug. I chose which algorithm families to ensemble.

Implementation belonged to the machine. Judgment belonged to me.
This isn't new. "Computer" used to be a job title. Switchboard operators manually connected every call. Typesetters spent years mastering metal letterforms. None of these jobs required general AI to disappear. They required standardization of the core output.

Jobs didn't disappear because machines could think. They disappeared because the core task got standardized. Cost collapsed. Employment contracted into a narrow premium niche. Survivors moved up-stack.

Data science is entering that cycle. So is coding. The question isn't whether AI will change your work.

The question is: at which layer do you choose to remain valuable?

I wrote about what this means for the rest of us:

Most of us make big life decisions with a hidden assumption: Somewhere out there is one correct path, and our job is to find it.

The right career. The right city. The right partner. The right next move.

It feels comforting, because it turns the chaos of possibility into a search problem. But it also keeps you stuck.

If you believe there’s one perfect path, every decision becomes terrifying. You overthink. You stay in roles, cities, or lifestyles that stopped serving you years ago, simply because they feel “safe”.

There isn’t one right path. There are many paths that could work.

The real question isn’t “What’s the correct answer?”  
It’s “Which of these viable options do I want to explore first?”

Practically, that has meant:
- Rating my current life honestly (health, work, relationships, joy)  
- Tracking my energy, not just my time  
- Talking to others such as retirees, founders, creators, people who already made big changes  
- Treating ideas as hypotheses and running small experiments instead of making giant leaps blindly

Life design isn’t about solving yourself once and for all. It’s about building the habit of continuous redesign.

Read more in my latest substack article:

Most projects don’t stall because the code is hard. They stall because we make the work heavier than it needs to be:
- We over-engineer the system.
- We hold ourselves to an undefined idea of “perfection.”
- Or we start coding while still mentally carrying other tasks and decisions.

None of this looks like a blocker on its own. But together, it’s enough to kill momentum.

Flow usually shows up when that extra weight is removed. Not when motivation spikes, but when the work is allowed to be simple and incomplete.

A few practical shifts that help when implementing a project:
- Decide what “working” means before you write code.
- Build the smallest version that proves the idea.
- Don’t refactor while you’re still getting it to run.
- Write down open questions so they don’t loop in your head.
- Leave cleanup for a separate pass.

Most of the time, progress doesn’t require more effort - just less interference.

Read more about this in my latest Substack article: