Build a data cleaning to executive summary agent suite from a written spec

Bronson Elliott, a data analyst for about 30 years, wanted his routine of cleaning, analysis, charts and write ups automated without coding. He first used an idea generator agent to turn the plan into a 713 line spec. Then on a Friday evening he pointed Claude Code at it. In about five hours it built five agents, for cleaning, statistics, Plotly charts, reports and short executive summaries, plus one that runs them in order on a messy CSV. It wrote 627 tests and fixed its own failures while he approved permissions about 20 times. A cost tracker put the build at roughly $35 in API usage. His main lesson is that the spec mattered more than the build.