One of the biggest AI unlocks I've found recently is having Claude do deep data analysis before working on features. You know, the type of analysis you always meant to do, but rarely did.
Now, it's easy... and every decision, even small ones, can be data driven with a minimum of effort. This means your work becomes much more targeted and impactful, immediately.
How? Give Claude Code a skill to write SQL against the production readonly DB. Since Claude Code runs inside your repo, it has access to your data models. Let it write SQL queries to temp files. Bundle the skill with a script that executes the given query against the readonly DB.
Now you can ask things like, "If I remove this branch of the function, how will that impact production traffic based on the last week of data?" Or, "We rolled out a change to classification last month. Please analyze its impact and suggest a plan to improve it further."
Codebase + production data + SQL + a really smart LLM = faster, smarter decisions, translated directly into production code.