Claude Code for Data Engineers: 12 Prompts, 10 Hacks, Fact-Checked

Everyone selling AI for data work quotes a number. I checked the numbers first, then built the workflow around what survives. The four studies on one screen: Copilot's famous +55.8% came from one toy task. The most careful study on real repositories found experienced developers 19% SLOWER — while believing they were 20% faster. DORA measured stability falling as AI adoption rises. Stack Overflow's top frustration: answers that are "almost right". Real gains exist, but they are conditional — this video is the conditions, built for data engineering. Inside: · 12 prompts for real data work — incident triage, SQL optimisation, safe migrations, data-quality tests, Spark tuning, backfills, legacy pipelines, PR review, cost audits · 10 setup hacks for data repos — CLAUDE.md, path-scoped rules, a hook that blocks destructive SQL, verification loops, subagents · Red flags: when NOT to trust the output 全部生产力数据逐条核实：12个数据工程提示词 + 10个 Claude Code 配置技巧。英文视频。