What Is an AI Data Agent?
Skippr is an AI Data Agent: a runner that helps teams extract, load, and model data into reviewable warehouse assets without hiding the output behind a proprietary runtime.
Like Codex, but for data. is the mental model. Codex reads a codebase, writes code, validates it, and gives you something to review. Skippr reads source metadata, syncs raw data, drafts dbt assets, validates them, and gives you warehouse tables plus a dbt project to inspect.
Why this category exists
Most teams already have sources, a warehouse, and a backlog of data work. The bottleneck is the repetitive setup between them:
- connecting sources
- handling schemas and drift
- loading raw tables safely
- scaffolding dbt models and tests
- validating the result end to end
That is the EL(T)M wedge for Skippr. AI Data Agent is the category. EL(T)M is the concrete technical job it performs.
What Skippr automates
- Extract: connect databases, files, and streams and sync raw data into your destination.
- Load: land bronze tables with incremental or CDC-aware behavior depending on the path.
- Model: draft silver and gold dbt assets, then validate them against the destination.
The important boundary is that ingestion correctness is not delegated to a model. Schema discovery, type mapping, incremental tracking, and CDC reconciliation are deterministic. AI is used where it accelerates reviewable modeling work.
Why this is trustworthy
- Reviewable output: Skippr writes standard dbt files and warehouse assets you can inspect and extend.
- Clear data boundary: row-level source data stays on the machine running
skipprand in your destination. - Scoped AI input: schema metadata is the default model input. Data samples are optional and off by default.
- Technical proof: CDC docs define order tokens, tombstones, and exactly-once final-state behavior instead of relying on generic automation claims.