Why I built ade-ops

Most data-platform tooling optimises for the happy path of a single product demo. Real analytics teams live somewhere messier: multiple environments, remote workspaces that are the real source of truth, promotions that have to be visible and reversible, and — increasingly — AI agents doing some of the work. ade-ops is my answer to operating that reality safely.

A few principles it's built around:

• The remote is authoritative, and every write is gated. You diff before you push; nothing changes a remote environment without an explicit confirmation. This is the single most important property when an agent is in the loop.
• One source of truth, many environments. You author once in src/; declarative overlays handle what differs between DEV, CERT, and PROD. No copy-paste drift.
• The framework admits what it doesn't know. Skills carry a maturity marker, and converters tag their output honestly — compat, light, heavy, or impossible. The last two are your call, not the tool's.
• Platform-indifferent by design. Databricks, Microsoft Fabric, or both — the engine treats each the same way. ade-ops isn't here to move you off one platform onto another; it's here to operate whatever you run.

It's open source, in early public preview, and built in the open. If you operate analytics platforms — with or without agents — I'd love your feedback. More notes to follow as the work continues.