OLD AI WAS BETTER

@programmergrind

12.7K followers · usually 2.7K views a post

Views

81.3K

Likes

4.5K

Comments

171

Shares

101

SoundTikTok sound · original sound - programmergrind

View velocity

+49.1K views over 10.2d · ~4.8K/day · 9 observations

Why it worked

The video taps into a common sentiment among developers about the perceived decline in code quality due to over-reliance on AI, presenting a relatable 'hot take' that sparks discussion and resonates with the target audience.

Summary

The creator argues that AI was better for coding a year or two ago, specifically mentioning ChatGPT's ability to refactor code snippets effectively. They contrast this with the current trend of using AI agents, which they believe leads to lower software quality due to developers becoming impatient and less involved in reviewing the code.

Structure

1Introduction: Old AI was better for coding
2Past experience with ChatGPT for code refactoring
3Current reliance on AI agents and its negative impact on code quality
4Comparison of developer involvement then vs. now
5Conclusion: Agents are powerful but often misused, leading to rushed and lower-quality code.

Product placement

ChatGPT (AI chatbot that generates code snippets and refactors code) - ACTS. Removing it would change the video's narrative about AI's impact on coding quality. Agents (AI agents that write code) - ACTS. They are presented as a new way of interacting with AI for coding, and their existence is central to the argument. Removing them would change the video's narrative.

On-screen text

OLD AI WAS BETTER

Original caption

hot take agents are incredible but it lowered software quality #Tech #ai #programming #computerscience #code

More from @programmergrind

More breakout videos from this creator.

Transcript

AI was better for coding about a year ago than it was today. The reason being that software quality was actually higher. When I used to code, I used to paste in snippets of my methods and then it came back to me refactored, my code. And I used to feel like the code was really good. And I did that the other day, again, instead of using an agent to actually go in and write code me, just sent a snippet of my code to ChatGPT. ChatGPT gave me back my code in a really nice way. And it made me realize that actually working with AI, working with AI two agents, back a year or two ago, when agents weren't a thing. Agents were a big part of building quality software. And a big reason why we're seeing prod incidents nowadays, left and right, cause people aren't reviewing their code before they, we were responsible, you were in the driver's seat for every single line of code. And although yes, you're moving slower and you deliver software longer, it takes you longer to reach deadlines. I do think paid quality, detail, and so you actually moved aggregated because you're introducing less bugs and you're more familiar with your codebase, you have more chemistry with it. Agents are powerful. I think it is the future of software engineering, it's not going anywhere. But I do think most software engineers misuse agents and they get impatient, they get nervous about deadlines and they try to move as fast as possible and they end up tripping over themselves 24/7.