Lately I’ve been doing a lot of coding using the AI Agent Claude Code. It is a terminal application which allows you to chat to it live while it attempts to code whatever you direct it to. My game Forward Blaster https://github.com/Sevii/forward-blaster was written by AI except for a few places where I edited constants. Needless to say the game includes a lot of loops. But I didn’t write any of them. I never felt the need to leave the AI agent workflow and start manually coding.
Which leads to the question ‘Have you written your last for loop?’ I don’t think I have but it’s very likely that time is coming. AI coding agents are quite good. You can tell them what type of iterator to use if you like. But in general it doesn’t matter a lot.
AI Agents also assist with debugging and architecture but not to the degree they do at writing code. LLMs can recognize many simple bugs and fix them without much human interaction. Where they really struggle is functionality that is not to specification. Even if you’ve given it the specification it cannot consistently recognize if something works but in the wrong way. And if there is a ‘one right way’ in this particular context your best bet is to tell the AI to use that ‘one right way’ specifically.
Software engineering is changing a lot. We used to write blog posts about minimizing toil so that engineers could focus on value add parts of the job. Few expected we could eliminate the time spent coding (the fun part) freeing up time for other engineering activities like planning, architecture, estimating, security review, oncall and the like.
Checkout my latest book Code Without Learning to Code on Amazon!