Hacker News
'Ask HN: How to Break Claude Code Addiction' Turns Into a Slot-Machine Argument With Concrete Countermeasures
A working engineer posted on August 30 that his days run long because he keeps prompting past the point of understanding the code he shipped, and asked how to stop. Multiple commenters independently landed on the same variable-reward framing, that a prompt tweak always feels like one more pull toward the jackpot. The actionable advice in the thread is specific: clear the session at 300k tokens and force the workflow to adapt, set a hard stop alarm and physically move the laptop, and spend the effort on architecture and chunking up front so the model works inside guardrails instead of producing output you have to review in bulk.
Source
↳ Follow the thread