What is ETL?

ETL stands for Extract, Transform, Load, the canonical pattern for moving data from operational systems into an analytics destination. You extract data from sources (databases, SaaS APIs, event streams), transform it (clean, deduplicate, join, reshape into a useful model), and load the result into a warehouse or other store. ETL transforms before loading, which suited an era when warehouse compute was scarce and storage expensive, you shaped data on a separate processing tier first. The modern variant, ELT (Extract, Load, Transform), inverts the last two steps: dump raw data into the warehouse or lake first, then transform it in place using the warehouse's own compute, often with tools like dbt. ELT has largely won for cloud analytics because cloud warehouses make in-place transformation cheap and because keeping raw data lets you re-transform when requirements change. A related pattern, reverse ETL, pushes modeled warehouse data back out into operational tools (CRMs, ad platforms) so business teams act on it. For agents, the relevant point is that by the time an agent queries data it is usually the transformed, modeled layer, and the freshness and correctness of that layer depend on the ETL/ELT pipeline behind it. Change-data-capture often feeds these pipelines to keep destinations near-real-time.