Designing Data Pipelines That Don’t Hate You Six Months Later

Stir Trek 2026 · AMC Easton, Columbus, OH · May 1, 2026

▶ Watch the recording · Slides — light (PDF) · Slides — dark (PDF) · Transcript Conference site · Session page at Stir Trek

Most data pipelines don’t fail on day one. They fail months later — the schema changed, a backfill double-counted half a table, the person who built it has moved on, and nobody can explain why last Tuesday’s numbers are wrong.

This talk is about the design choices that decide which of those futures you get: schema evolution, idempotency, observability, testability, and designing for change. None of it is a slogan. It’s a set of concrete decisions you make early, before the pipeline has any data worth breaking.

My first conference talk — part of the Stir Trek 2026 playlist.

Sadukie (Sarah Dutkiewicz) wrote it up in the NimblePros Stir Trek 2026 recap.