Sentry Co-founder: LLMs Slow Long-term Development Speed, OpenClaw's Excessive Code Generation Makes It Hard to Recover

Sentry co-founder David Cramer posted on X recently, stating he is 'completely convinced' that large language models (LLMs) are not currently a net productivity gain. While LLMs lower the entry barrier for development, they continuously generate complex and hard-to-maintain code, slowing long-term development speed.

Cramer questions the 'agentic engineering' approach—letting models generate code automatically and deploy directly—arguing that the code quality is significantly worse, and accumulated code becomes a net burden. Specific issues include poor performance in incremental development of complex codebases, interfaces not conforming to language conventions, and 'pure slop test generation'. He specifically named OpenClaw, saying such tools are 'irrecoverable' due to excessive code generation.

Cramer's judgment is based on his experience with mature codebases. His recent increased contributions are because it's 'fun', not because development has become easier; there's no essential difference in actual time spent.