Mori LiuProduct & Growth @ Hume AI | Creator (120k+) | prev. AI/HCI Research @ Columbia

recent reflections on #ai #coding

Name: recent reflections on #ai #coding
Uploaded: 2026-10-19T16:00:31+00:00
Duration: 1 min 22 s
Channel: Mori Liu
Description: In this solo reflection, Mori Liu argues that merely building AI‑powered applications does not make one an engineer and that learning to code from…

19 Oct 2026 · 1 min

Watch on YouTube

1:22

In this solo reflection, Mori Liu argues that merely building AI‑powered applications does not make one an engineer and that learning to code from scratch is not the solution for modern technical work. Drawing on his CS background and experience releasing software, he references an Anthropic video on “vibe coding,” likening AI‑assisted coding to the historical impact of compilers on assembly language: both enable rapid scaling, but unlike deterministic compilers, AI remains nondeterministic. Consequently, being technical today is less about writing code and more about effective technical management—acting like a CTO or PM who deliberately assigns “leaf node” tasks (independent end‑level features) to AI agents, thereby improving scope clarity and verifiability. He stresses the importance for leaders to understand system components such as libraries, database schemas, and architecture, even when AI performs the work, drawing an analogy to knowing how to clean a house before hiring a cleaner. The talk emphasizes intentional task design, verification, and deep system literacy as key competencies in an AI‑augmented development landscape.

Chapters

Being able to build apps with AI doesn't necessarily make you an engineer, but I don't think the solution is learning coding from scratch. So let me explain. I personally got a CS degree, but I don't mainly work as a software engineer. But after building and releasing several software products on my own, I didn't feel like my engineering ability was increasing compared to the days where I was implementing BERT from scratch or writing C++, which led me to this video released by Anthropic talking about how they vibe code in production, where the speaker makes a very interesting analogy of what vibe coding is doing to software engineering is very similar to what compiler did to assembly code. Initially, engineers didn't trust it, But over time, they did realize using compilers allowed them to move very fast at scale. However, how AI is different from compilers is that AI is not completely deterministic. No matter how intelligent the model might get, there's still a level of unpredictability. So my take on how to be technical in this era is really how to become a better manager. So it's more like a relationship of a CTO or PM to an engineer, where number one, you have to be intentional in task assignment. In the video, they call this leaf nodes, which are end-level features where no code is dependent on them. This gives the agent a reasonable scope, but more importantly, it increases verifiability. And finally, I think it's extremely important to understand the components of a system because it's very different to hire a cleaner to clean your house when you know how to clean but you don't want to do dirty work versus you don't know how to clean at all. So it's extremely important to extrapolate technical components like a new library or database schema and really understand yourself, even though it is the agent that's carrying out the task.