ETL fragmentation shifts to observability
Conor McCarter, co-founder of Prequel, on Fivetran's existential risk
The integration layer is fragmenting faster than any one ETL vendor can unify it. In practice, teams pull data from a mix of Fivetran, Airbyte, vendor built warehouse exports, cloud native sharing features, and hand rolled jobs, so the monitoring problem expands beyond one connector catalog into the whole pipeline. That is why the natural control point shifts from ETL itself to observability and workflow tooling that watches many systems at once.
-
Fivetran won by building a relatively small set of high trust connectors and charging a premium to keep them working as source APIs change. That model creates quality, but it also makes universal coverage expensive, because every new connector adds ongoing maintenance work, not just initial build work.
-
Airbyte attacks the long tail with community built connectors, which broadens coverage but makes reliability less uniform. That split shows why one tool rarely becomes the single dashboard for everything. The market keeps separating into curated connectors for critical systems, and broader but noisier coverage elsewhere.
-
Warehouses are building tighter native sharing inside their own ecosystems, not a cross warehouse command center. Snowflake Secure Data Sharing shares selected database objects between Snowflake accounts, and Redshift datashares share live data across Redshift clusters and workgroups. Both deepen same stack lock in rather than aggregating all connectors everywhere.
The next step is a stack where important SaaS vendors ship their own warehouse feeds, ETL vendors focus on the hardest and most widely used connectors, and monitoring moves into dedicated data observability layers. That makes the winning product less likely to be one universal ETL pane, and more likely to be the system that explains where the pipeline broke, across every ingestion path.