Databricks + Cursor · Live job triage
Cursor closes the loop on Databricks job operations — Slack trigger, plan-first investigation, Spark UI forensics, notebook patch, Codex-reviewed PR.
Built for a Databricks AE pitching senior data engineers stuck context-switching between VS Code, the Databricks UI, and Spark UI. Click through one bleeding job’s full ~82-second scripted lifecycle: incident → planner → 5 specialists → synthesis → patch → cluster validation → ship.
Slack-to-PR closed loop
The agent reads the same Spark UI, run history, cluster, code, and lineage your senior engineer would, then replies in the original Slack thread with a merged PR.
Planner-orchestrator
One Opus planner intakes the incident; five Composer specialists (skew, cluster, code, lineage, cost) dispatch in parallel; the planner synthesizes a ranked 4-lever remediation.
Plan visible before execution
Stakeholders see how the agent will investigate — a reproducible runbook, not a black-box fix. Plan-first is the durable differentiator.
Guardrails every remediation PR must clear