Hacker Timesnew | past | comments | ask | show | jobs | submitlogin

Imagine if you will, a team tasked with building widgets. That means designing the widget, the mechanisms to fabricate widgets from input materials, and building those mechanisms to then build the widgets.

You could do this with a team of two people, the engineer and machinist. The engineer spends their time at the drawing board and handing designs to the machinist to fabricate. The machinist is probably going to run into problems, or there are problems with what is fabricated and the engineer needs to go back to the drawing board. It's an iterative process.

Now imagine those widgets are software. Historically we call both the engineer and machinist "software engineers" or "developers" and might call the first one "senior" or "architect" or something like that. Since we don't need a human being to physically operate machinery it seems like using AI agents would be the an ideal place to replace the software machinist.

However there's a problem - the engineer/machinist duo works not because the machinist is an engineer but somehow worse; they have different skillsets and the engineer is going to get immediate feedback from the machinist that a design is going to work or not. The engineer/machinist divide works because the machinist can say "nah."

AI agents today don't have that ability. You can tell them to do things that are impossible, or they will make shit up. Without the ability to check for validity or preemptively avoid mistakes, vibe coding is just shitty software development. If and when that problem is cracked then the concept of a software machinist will be obsoleted.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: