DAILY BRIEFING · SUNDAY, JUNE 7, 2026
A quieter post-Summit weekend, with the through-line clear: open table formats and the catalog are hardening into the control plane for AI agents — DuckDB pushes read/write parity across Iceberg, Delta and Lance, Supabase bets on horizontally scaled Postgres, and Unity Catalog, BigID, OpenLineage and the new MCP servers race to govern what agents can touch.
⇣ Jump To
Table Formats · Specialty Platforms
Governance, Security & Compliance · Data Contracts & Lineage
⚡ QUICK TAKES
| Story | Signal |
|---|---|
| ↗ DuckDB-Iceberg v1.5.3 adds MERGE INTO and catalog writes | The single-node engine becomes a full read/write Iceberg participant, not just a reader. |
| ↗ Single-node engines (DuckDB, DataFusion, Polars, LakeSail) mature | Out-of-core, lakehouse-native transforms keep eroding the case for a cluster. |
| ↗ DuckDB Delta gains writes, Unity Catalog and time travel | Format-agnostic read/write across Delta and Iceberg is becoming table stakes. |
| ↗ Lance lakehouse format gets a DuckDB test drive | Multimodal/AI-native table formats are entering the mainstream query path. |
| ↗ Supabase raises $500M at $10.5B, previews Multigres | Horizontally scaled Postgres is being positioned as foundational AI-app infrastructure. |
| ↗ Snowflake Intelligence + Cortex Code become an agentic control layer | Consumption is shifting from dashboards to governed, in-platform agent execution. |
| ↗ Databricks Genie ships a Managed MCP Server (Beta) | Any agent can now query Genie Spaces + Unity Catalog in natural language over MCP. |
| ↗ Governing AI agents at scale with Unity Catalog | Identity-aware, runtime policy enforcement for agents, models and MCP servers in one plane. |
| ↗ BigID named a Forrester Wave Leader, extends into AI security | Classification platforms are repositioning around agent access governance and DLP. |
| ↗ IBM backs OpenLineage for structured + unstructured lineage | Explainable AI needs lineage that spans documents and tables, not just pipelines. |
DuckDB · May 2026
The v1.5.3 Iceberg extension adds MERGE INTO support and catalog-backed writes, turning DuckDB from a read-mostly client into a full participant in REST-catalog-managed Iceberg lakes (Polaris, Nessie, S3 Tables). For data engineers, that means delta-style upserts and table maintenance can run from a local or embedded process without a Spark cluster in the loop — and the same engine reads Delta and raw Parquet alongside Iceberg.
✍️ Tom Ebergen & Thijs Bruineman · Read article →
DEV Community · May 2026
A survey of the single-node engine landscape argues the gap to distributed compute has narrowed sharply: Polars stabilized its streaming execution engine for out-of-core datasets and added direct sink_iceberg/sink_delta cloud sinks, while DataFusion and LakeSail round out a Rust-and-Arrow stack that reads and writes lakehouse tables natively. The practical takeaway for platform teams is that a large share of transformation workloads no longer justify a cluster — and the cost math is shifting accordingly.
✍️ Alex Merced · Read article →
DuckDB · May 2026
DuckDB's Delta extension moves beyond read-only: it now supports writes, Unity Catalog integration, and time-travel queries against Delta tables. Paired with the v1.5.3 Iceberg work, it puts read/write parity across both dominant open table formats inside one embedded engine — and lets DuckDB resolve tables through Unity Catalog rather than raw paths. For architects, that strengthens the case for catalog-mediated, format-agnostic access as the default pattern.
✍️ Ben Fleis · Read article →
DuckDB · May 2026
A joint LanceDB/DuckDB write-up walks through querying the Lance format — an AI-native, columnar table format optimized for multimodal data and vector workloads — directly from DuckDB. With LanceDB's mindshare growing faster than any peer this year on the strength of serverless and multimodal use, native engine access signals that AI-oriented table formats are moving from niche vector stores into the general analytical query path that data platform teams already operate.
✍️ LanceDB team & Guillermo Sanchez · Read article →
CNBC · June 2026
Supabase closed a $500M round valuing the Postgres-as-a-service company at $10.5B, riding the vibe-coding wave, and previewed Multigres — a horizontal-scaling layer (Vitess-style) aimed at letting Postgres deployments scale "up to the size of OpenAI or larger." The signal for architects: rather than migrate operational workloads to a bespoke distributed database, the market is doubling down on scaling Postgres itself — the same engine that already underpins pgvector and a large share of AI-app backends.
✍️ CNBC · Read article →
Techzine · June 2026
Snowflake expanded Snowflake Intelligence and Cortex Code to act as the control plane for agentic work, with Cortex Code now reaching external systems including AWS Glue, Databricks and Postgres. The shift to watch is architectural: consumption is moving from human-driven dashboards toward governed agent execution that runs where the data lives, which puts more weight on the semantic model and access controls beneath the natural-language surface.
✍️ Techzine · Read article →
Microsoft Learn · June 2026
Genie now exposes a Managed MCP Server in Beta, letting any MCP-aware agent ask natural-language questions across Genie Spaces and Unity Catalog data and receive grounded answers. Because the retrieval grounds on Unity Catalog's semantic and governance layer rather than a separate YAML model, it ties agentic data access directly to the catalog — reinforcing the pattern where the metadata layer, not the BI tool, becomes the entry point for AI consumption.
✍️ Microsoft Learn (Azure Databricks) · Read article →
Databricks · June 2026
Databricks details how Unity Catalog and the Unity AI Gateway provide a single governance layer for AI agents, models, MCP servers and data — enforcing identity-aware access, runtime policies, guardrails and full auditability across every agent interaction. As agents make orders of magnitude more data requests than humans, the argument is that governance has to move from static table grants to runtime, per-call policy. For platform engineers, it frames the catalog as the enforcement point for agent activity, not just a discovery surface.
✍️ Databricks · Read article →
BigID / PR Newswire · May 2026
BigID was named a Leader in The Forrester Wave: Sensitive Data Discovery and Classification, Q2 2026 (one of three Leaders, with a 5/5 across eleven criteria), and announced four AI-security capabilities: DLP Prism for context-aware loss prevention, AskBigID GPT for natural-language posture queries, Agentic Access Governance for controlling what AI agents can reach, and Integrated Employee AI Governance for sensitive data flowing through employee AI tools. The direction echoes the week's theme — discovery-and-classification vendors are repositioning around agent access as the new control point.
✍️ BigID · Read article →
IBM · May 2026
IBM is backing OpenLineage to capture lineage across both structured tables and unstructured content, arguing that explainable AI requires tracing how documents — not just rows — flow into model outputs. With OpenLineage already the cross-tool standard emitted by Airflow, Spark, Flink and dbt, extending it to unstructured assets pushes lineage toward the RAG and agent pipelines that increasingly drive enterprise AI, where provenance has been hardest to establish.
✍️ IBM · Read article →