ClickHouse at 10, dbt's AI push, and new cloud connectors

A decade of open source, a summer of connectors

ClickHouse turned ten last week — officially open source since June 15, 2016 — and the team published two reflective posts on what that decade looks like in numbers and community. But the birthday presents came in the form of real product expansions. Postgres Managed by ClickHouse, which entered public beta in May, now ships RBAC, a Terraform provider, ClickPipes integration, and PostgreSQL extensions. That's a deep set of enterprise features for a service that's been in beta only a month.

On the ingestion side, ClickPipes gained a GCP Pub/Sub connector in private preview, letting teams stream Cloud Logging, GKE events, and custom GCP application signals directly into ClickHouse without routing through Cloud Storage or Dataflow. And Tinybird's Forward added a DynamoDB connector that mirrors an entire table via Point-in-Time Recovery snapshot plus CDC from DynamoDB Streams — another low-friction path from operational data stores to real-time analytics.

dbt's agentic data stack, from theory to alpha

dbt Labs published three guides on building agentic data pipelines, making those pipelines trustworthy through the transformation layer, and a new skill called "context engineering" for structuring information so AI agents can reason reliably. They also shared a migration guide arguing teams should "start fresh, don't lift and shift" — though that advice feels aimed at organizations still on ETL tools rather than the agentic future of the other three posts.

On the engineering side, dbt-core v2.0.0-alpha.2 shipped with several dbt Fusion additions: Homebrew installation via brew install dbt, --empty support in the seed command, and a GET /api/v1/models endpoint that includes warehouse catalog statistics (row count, bytes, last modified) alongside the materialized field — feeding dbt build --write-index. There's also dbt-core v1.12.0b3 with a direct_parents attribute for model nodes, letting lineage consumers render DAG edges from nearest public ancestors rather than the full dependency graph.

Bugfixes and ecosystem thinking

Dagster 1.13.10 fixed two sticky bugs: stale materialization events surviving asset wipes, and asset backfills getting stuck without emitting further runs. The ecosystem post from ClickHouse isn't a release per se, but it's worth noting as a companion to the ten-year anniversary — it catalogs the clients, dashboards, pipelines, projects, and now AI agents that have grown around the database, a reminder that platform adoption is measured as much by what builds on top as by what ships inside.