Build vs Learn

An opinion piece

The following is my own opinion, not my employer’s.

In times when we did not have coding LLMs, my usual routine of learning on the weekends and during some evenings was to:

1. come up with a sufficiently interesting / challenging task
2. sketch a solution (in a paper notebook)
3. decide on the programming language
4. start coding
5. probably fail
6. rinse and repeat
7. either abandon the idea or complete a POC

The 4–6 part was the most time consuming, sometimes several evenings. But you could also say this was the time when I’d learn the most. These learnings could be:

• a particular framework I depend on does not build correctly (at all or on my machine)
• I don’t fully know, how to build what I’m trying to do
• I got down the rabbit hole of trying to master a completely new area with a steep learning curve (like coding an iOS app — Android was way easier for my Java background)

By the end of this exercise I would either have a working prototype, that I could ship for others to see, or just learnings and bruises. Both were important in their own ways, as part of a discovery process.

These days, my coding (can I call it coding at all?) routine has changed to:

1. come up with the idea (became harder, I don’t know is it age or just lack of training I described above)
2. plan with a coding agent
3. let it build it all
4. check the result
5. rinse and repeat
6. rarely fail
7. get the POC up and running

The only somewhat interesting or somewhat challenging part is 2: this is where I push myself to explore the edges of the idea and make sure the agent does not come up with a shortsided implementation. Every other step abstracts me from the logic of code as much as possible. This is where I begin to miss the touch with my idea with getting into the weeds of the implementation. This part is crucial for making sure that I get the necessary feedback to my idea. Because it may be too shallow, or too abstract, or too ambitious.

By abstracting the code creation, LLM forces me stay in the confines of the idea, rather than see and feel the hurdle of implementing it into life. It is like calling someone to study for you to pass an exam. I know it is the extreme analogy, but does convey the feeling.

To mitigate this, I began changing my routine by:

• looking for and reading long complete texts (thanks to Hacker News for being the tresure trove of these), not just announcements with lots of likes on social media
• reading books, including technical ones, like this one
• spending longer on my idea away from an LLM (preferrably away from my keyboard too to avoid the temptation of typing into the LLM input right away)
• writing without LLM (like this text)
• watching longer thoughtful videos (examples: 1, 2, 3)

What’s your way of learning today? Did it change with the advent of LLMs?