Cüneyt Mertayak
← All articles

Learning

On Top-Down and Bottom-Up Learning

A wooden arch assembled from blocks, with separate foundation and wedge-shaped blocks beside it.

Early in my career, I was about to contribute to a codebase built with Ruby on Rails. I didn't know Ruby or Rails, so I began learning the language and framework before starting the task.

At the time, that seemed sensible. How could I contribute without understanding what the codebase was built on?

Looking back, I think I could have approached it differently. I already knew several programming paradigms and understood concepts such as variables, functions, and control flow. Although Ruby's syntax and Rails' conventions were new to me, I wasn't starting from zero. I could have followed the code related to my task and looked things up as I encountered them. I might have spent less time learning parts of Rails that I didn't need for that contribution.

That experience made me reconsider a habit I'd had for a long time: trying to understand a subject from the ground up before doing anything with it.

Since then, I've paid more attention to where I start. Sometimes I need the foundations first. Other times, starting with a task helps me see which foundations are worth learning. That distinction has helped me approach unfamiliar work, and I want to share how I think about it.

Two ways into a subject

I think of starting with a problem or goal and then learning the details that help me make progress as a top-down approach.

Someone who learns programming to mod Minecraft is a good example. They might start by changing something in a game, then learn enough code to understand why it works or why it doesn't. Their goal gives them a reason to explore.

My habit has been more bottom-up. I like learning the pieces first, then putting them together to understand the bigger picture. Books and structured education often follow this order. A programming book may introduce variables, data types, control flow, and functions before showing how they come together in a project.

I find that reassuring. When I understand the foundations, I feel more confident reasoning through a problem. That knowledge also carries over: my earlier programming experience gave me a way into Ruby, even though I hadn't used the language before.

The same choice comes up outside programming. Whenever I'm learning something unfamiliar, I can start with the foundations or begin with a question and learn my way toward an answer.

When learning becomes a delay

The trouble starts when I treat understanding the foundations as a requirement before taking any action. A subject can be large enough that there is always something else to read.

I've also seen people around me collect books they never read or start tutorials they leave before the ten-minute mark. I don't know why each person stopped. But I wonder whether some of that material would have held their attention longer if they had first tried something and discovered a question they wanted answered.

For my Rails task, the distinction matters because I had a specific contribution to make. Learning the entire framework was a much larger undertaking than learning enough to begin that contribution. Starting with the task could have made it easier to decide what deserved my time.

Moving between the two

I don't want to give up bottom-up learning. I want to be more deliberate about when I use it.

When I'm curious about a new subject, I can try something small first. If the experience holds my attention, I can study the foundations and return to it with a better understanding.

When I have a concrete task and limited time, I can start by investigating the part closest to the task. If I keep getting stuck on the same concept, that's a sign I should step back and learn it more deeply.

And sometimes depth is the goal from the beginning. In that case, working through the fundamentals may be exactly how I want to spend my time.

Learning with AI

AI tools have given me more ways to explore a subject before deciding how deeply to study it. I can ask an LLM for an overview, use a research agent to find sources and questions I hadn't considered, or ask a coding agent to help me find my way around an unfamiliar codebase.

If I faced the Rails task today, for example, I could ask a coding agent to trace the part of the code related to my task and explain what it finds. That could help me get started sooner, while also showing me which parts of Ruby or Rails I need to understand better.

These tools can help me navigate, but I still need to check their work and develop enough understanding to judge it. They make it easier to move between exploring a subject and studying its foundations.

When I want to start by doing something, I might ask:

I want to explore this subject through a practical task. Suggest a small place to start, give me an overview of the ideas involved, and help me learn the details as I need them.

When I want to build from the foundations, I might ask:

Teach me this subject from its foundations. Start with the basic definitions and assumptions, explain each step with simple examples, and suggest a path from those foundations to the larger ideas.

I've used this distinction in the years since that Rails experience. When I face something unfamiliar, I ask myself: What am I trying to do, and what do I need to understand to take the next step? Sometimes the answer is to study the foundations. Other times, it's to begin with the task and learn along the way.