How to Maintain Technical Judgment When Using AI-Assisted Programming
This is quite interesting.
You all know Karpathy, right? The AI heavyweight whoâs worked at both Tesla and OpenAI. He recently dropped a bombshell in the developer community: âAfter coding for 20 years, using AI now feels like cheating.â
Honestly, I laughed when I first read that. Isnât this the exact dilemma we face every day? When using Copilot, weâre simultaneously thrilled (âHoly crap, it can auto-complete this?!â) and guilty (âWho actually wrote this code?â).
1. From âVibe Codingâ to âFeels Like Cheatingâ
Karpathy previously introduced the concept of âvibe coding,â which roughly means letting AI sense your programming intent and semi-automatically write code. At the time, everyone cheered, calling it the future. Now, heâs backtracking, saying the model has issues.
This is painfully relatable. When AI-assisted programming first emerged, everyone was hyped. But once the novelty wore off, the real problem surfaced: Efficiency improved, but the sense of accomplishment from coding vanished. Itâs like using cheats in a gameâyou finish faster, but itâs utterly unsatisfying.
2. That 110-Task Experiment
The article mentions a wild experimentâan AI Agent independently completed 110 coding tasks, including hardcore stuff like setting up a home surveillance system. Sounds impressive, right? But anyone in tech knows: Getting it to run and making it production-ready are two different things.
Iâve seen too much AI-generated code:
- It runs, but itâs spaghetti code thatâs a nightmare to maintain.
- Zero commentsâthree months later, even its creator wouldnât recognize it.
- Crashes at the slightest edge case.
So hereâs the reality: AI can handle 80% of the boilerplate code, but humans still have to clean up the remaining 20% of devilish details.
3. What Are We Really Afraid Of?
Saying it âfeels like cheatingâ reveals programmersâ deeper anxieties:
- Fear of becoming âAI code proofreaders.â
- Fear of losing technical judgment (just blindly following AIâs suggestions).
- And the biggest fearâwhat if bosses think, âIf AI can write code, why do we need you?â
But hereâs a different perspective. When cars replaced horse-drawn carriages, carriage drivers became drivers. Now that AI handles repetitive tasks, shouldnât we focus on cooler things? Like designing architectures, optimizing algorithms, or solving complex problems AI canât handle?
One Final Observation
The Chinese communityâs take on this discussion is uniquely pragmatic. While Western debates revolve around âWill AI replace programmers?â, weâre more concerned with âHow to use AI to stay competitive.â Maybe thatâs the difference between practitioners and theorists?
In the end, Karpathyâs statement is a wake-up call for everyone: The more powerful the tool, the more important it is to know when not to use it. Just like you wouldnât forget that 1+1 equals 2 just because you have a calculator, right?
(The End)