Agentic Data Science in Practice: Agents Write Code but Answer Wrong Questions
hugobowne · x · 2026-07-27
The PyMC Labs team shared insights from building agentic data science workflows over the past few months, covering agentic EDA, causal modeling, autoresearch, and layered verification.
They found that writing code was rarely the hardest part. While agents can explore data, build predictive pipelines, and critique each other's code, they exhibit critical flaws:
- Producing code that runs perfectly but answers the wrong question
- Making confident claims unsupported by the data
- Missing data leakage
- Losing important context during longer workflows
To understand the field's current state, they launched the State of Agentic Data Science 2026 survey, asking practitioners where agents genuinely improve data work and where they fail.
More from coding & agent
- Tokyo Agent Forge hackathon shipped production-ready AI agents in one day — DavidBennett__ · 2026-07-27
- Long-running agents will need immutable event logs, this thread argues — sebpaquet · 2026-07-27
- A VS Code extension adds Markdown-style highlighting to Alchemy string templates — samgoodwin89 · 2026-07-27
- Claude’s Stripe MCP connector is being called unusable after repeated disconnects — evielync · 2026-07-27
- Reddit user says Vecel’s Eve looks like “Next.js for agents” — richie9830 · 2026-07-27
- Leader.skill open-sources a framework for turning vague requests into executable agent goals — 数字生命卡兹克 · 2026-07-27