I’m sick of AI “Thinkslop” in my PRs

Short summary

A developer with a mechanical engineering background argues that single-model AI code generation produces "Thinkslop" — code that looks correct but contains hidden logic flaws. They advocate running code through 8 different LLMs simultaneously and treating disagreements between models as signals for where real bugs hide. The author built a local-first tool called NexaVerify to orchestrate this multi-model consensus approach and found bugs that individual models consistently missed.

• •Single-model AI code generation can produce plausible-looking but logically hollow code ("Thinkslop")
• •Running code through 8+ models simultaneously surfaces bugs where models disagree on architecture or logic
• •Author built NexaVerify, a local-first multi-model consensus engine that caught bugs individual models missed

Generated with AI, which can make mistakes.