The Coming Loop
Points and comments are a snapshot, not live.
Coding agent loops are reshaping software development, creating codebases that humans may no longer understand.
The author describes a pattern where work is queued for an AI agent, which attempts it, and then a harness loop decides if the task is done. This creates code that is defensive, complex, and avoids strong invariants. Loops work well for porting code, performance exploration, and security scanning, but produce code the author finds distasteful for long-lived systems. The author worries about losing the ability to understand code, creating cognitive dependencies, and the pressure to adopt loops built by attackers and competitors.
What commenters are saying
Many commenters share the author's unease, with some reporting that AI has killed their passion for coding. One commenter describes feeling like they spent years transcribing a phonebook next to a photocopier. Another states they are "at the point where I think it's dumb to not do it but also dumb to do it." A third commenter finds value in using AI for small personal projects that were previously not worth the effort. Some defend the analogy to early automobiles, arguing it's too early to judge the technology's eventual impact.
A separate camp debates the inevitability and desirability of this shift, with some pointing to a $250M valuation for a one-person startup (Polsia) as a sign of progress, while others question whether speed and profit justify the loss of understanding.