MCP or CLI? Why I Use Both for AI Data Analysis
Explore data with MCP, review the SQL, and save a CLI command to run it again. A practical AI data analysis workflow on a Mac.
Read the noteField notes
Explore data. Explain the result.
Practical notes on working with AI assistants, asking better SQL questions, and investigating data changes. Worked examples connect a question to the calculation and the evidence behind its answer.
How AI can investigate a 12 GB export without reading it into a prompt
A worked example of data grain, two-stage SQL aggregation, and Parquet scan planning when an AI assistant helps investigate a large local export.
Read the noteThe AI agent checked the table. Did every result describe the same state?
Why agent data tools need exact table identity, bounded coverage, and honest incomplete results beyond a successful tool call.
Read the noteThe report says verified. Is the change safe to release?
How to keep table identity, evidence coverage, and release intent separate in a data investigation report.
Read the noteThe query succeeded. Which table state did it read?
Why a Delta or Iceberg table is not the same as every Parquet object in its storage location.
Read the noteThe pipeline passed. The data still needs proof.
How to compare two data states and leave evidence another engineer can reproduce.
Read the note