AI-built everything, and a growing allergy to the slop

OpenTPU is an open-source AI accelerator developed by AI, following earlier RISC-V cores built the same way. It reportedly runs Qwen 3.5 and Gemma on an FPGA board costing around $300. A commenter on a shader component library said LLMs work well for niche, formerly inaccessible kinds of programming where imperfect-but-good output is fine. Someone built a media library with Claude "this week" for their own GPU outputs. A 60-year-old on Ask HN says Claude Code reignited their passion for building.

The backlash runs in parallel. Comments on the ESP32-C3 adblock call the docs "AI vomit." AstroHelm got a bug report for what looks like AI-generated copy. An Ask HN proposes a flag for AI-generated articles, and others note HN's guidelines already bar generated text in comments. A "Forever Junior" essay argues that getting an AI to cooperate is a different skill from the craft of coding.

This is the key bit: the code is getting cheap, and trust is the thing that isn't. People forgive AI-built tools when the human clearly steered them and tested the result. They get angry when the README or the claims feel unreviewed.