Everybody Guesses
Hi Reddit. This isn't specifically a Ruby or Rails post, but it's about something anyone maintaining a mature Rails application will recognize.
Code tells you what happens without telling you why anybody wanted it. We've always filled that gap by reading the code, guessing the intent, and asking somebody when the guess felt wrong.
AI coding agents do the same thing, except they can make a great many reasonable guesses before anybody notices one was wrong. The code still works. The tests still pass. The guess becomes one more pattern in the application, which makes it look even more intentional to the next person or agent.
I've spent a couple of decades living with the consequences of my own good ideas, so I wrote about why this worries me. The post connects agentic coding to Peter Naur, Tony Hoare, XP, software factories, and the old-fashioned responsibility to say no to code that works but is too difficult to understand.
It also argues for using faster agents to take smaller steps. More research, more feedback, and less archaeology at the end of a giant pull request.