Don't Skip Learning!

This is a modified blog post of some thoughts I shared with my colleagues at Meta in April. With the agentic coding boom well on its way, I placed a new pressure on myself to do even more: parallelize everything, ship more code, just throw AI at it. This is not the way. This post is a reminder to myself that comprehension is paramount. Comprehension opens doors, is the root of vision and is necessary for long-term success.

tl;dr – It's never been easier to learn something new or to go incredibly deep on a topic or technical area. Don't tradeoff learning for moving fast.

I was reflecting on the first bit of code that I'd shipped at Meta, the code created a schema for a new logging event. I was very new, so I copied the boilerplate of another event I knew was functional and ran some auto-generated tests and lints to prove that what I'd copied and edited was functional. Probably quite similar to what claude would do today if I threw it at the same task. My mindset in that moment was "ok, reference these existing implementations and we're probably good here. I don't fully understand all of this yet, but it's functional. I'll have to read more about this later." I was concerned about moving fast and wasn't yet sure where learning fit into my workflow.

But if I were working on this same task for the first time today, my approach would be quite different. Now the time it would take me to copy-paste boilerplate and tweak some fields would be negligible, instead I'd be able to use most of the time to ask questions, rather than choosing to delay my learning.

I would ask about the function of certain class variables, the relationships between components, how terms in the documentation relate to the concrete implementations, the testing framework, what a basic test guarantees about the code, how what I wrote interacts with other systems in the broader architecture.

Learning on demand is a cornerstone of working in tech. There have always been new frameworks and new tools that have made continuous education an expectation of working in this industry. I've personally always found this to be a highly engaging and fun part of the job. I think we can lose sight of the joy of learning when we see it as an obstacle to completing an urgent task, but it's incredibly important to continue to invest in your own comprehension, particularly now with the volume of code absolutely surging y/y.

There are real concerns that individuals' skills are declining due to the use of AI, but I wanted to remind you (and really, myself) that improving your baseline technical skill is more important now than it has ever been. The good news is that it's never been easier to learn something new or to go incredibly deep on a topic or technical area. If you consider yourself a junior engineer, then this is doubly true for you!

I'll close by highlighting a January study from Anthropic, which opens with a bold claim: "We found that using AI assistance led to a statistically significant decrease in [coding] mastery." That's alarming, but if you read on, there's an important detail. Outcomes were different depending on how the participants used AI. Those who use AI to also improve their comprehension performed better. AI didn't prevent learning, it just made it easy to skip. Don't skip learning!