5 picks from Hacker News, Lobsters, dev.to & Mastodon · No slop, no self-promo
Developer reports sudden degradation in Claude Opus performance on a growing codebase — but no specifics on what broke or how to reproduce it.
Genuine frustration-driven question with no technical depth or specifics. No code examples, no reproduction steps, no context about the project. Reads like venting rather than seeking actionable help.
Developer corrects their own token tracking tool to account for cache costs — reveals that 99% of tokens are cheap cache reads, not expensive input tokens, with real dollar figures ($363/5hrs).
Iterative refinement of a real measurement tool with honest correction of previous methodology. Shows actual cost math and specific numbers ($363 per 5 hours), not marketing.
Subreddit megathread for collecting Claude usage limit complaints and workarounds — community organization, not a vibecoding technique.
Meta post organizing community discussion, not a vibecoding technique or tool insight. Useful for the subreddit but not actionable for developers building with AI assistants.
Inflammatory headline with no supporting evidence or substance — pure engagement bait.
GPU hardware comparison question — not relevant to AI-assisted coding workflows.
Missing something good? Send it our way.