Agents now have to earn their keep in data

Snowflake made the clearest positioning turn. On 4 Nov, its homepage hero changed from Code Work to The Agentic Enterprise Runs on Snowflake, with new copy around trusted enterprise AI, governed data, controlled actions, models and clouds. The same day, Snowflake announced AI platform additions including Business Ontology, intended to improve AI agent reliability, and autonomous bots for its CoCo and CoWork AI agents. On 8 Nov, it added a free global training programme and Snowflake CoCo for Excel and Google Sheets.

Sigma moved in the same direction from the application layer. On 3 Nov, it added new AI-agent blog content and nine product pages, including agentic analytics and use-case pages. On 4 Nov, its navigation changed to include Agents, MCP & AI Tools, Agentic Tools, Data Models, Connected Spreadsheet, and Python / SQL, while removing entries including AI Toolkit, Finance, Supply Chain and Healthcare. That is not a small copy edit. It is a map of where Sigma wants the buyer to start.

The timing matters. Snowflake began a new Google ads push with 14 creatives observed on 2 Nov, with earliest first shown on 1 Nov. Further Snowflake creative changes were observed on 5 Nov and 7 Nov, with at least 40 creatives running by 7 Nov. In the same week, Sigma reworked its navigation and product pages. When two companies at different layers of the stack change the front door within 48 hours, the category is not just talking about agents. It is reorganising merchandising around them.

Databricks had the broadest infrastructure week. On 30 Oct, it announced general availability of Real-Time Mode for Spark Declarative Pipelines and smarter cost controls for serverless. On 1 Nov, reports said it expanded its SAP partnership with three products: Genie, Lakeflow and Lakebase. On 6 Nov, it introduced Apache Spark 4.3 as part of Databricks Runtime 19, and Lakehouse Postgres became generally available on Google Cloud.

The ecosystem build-out was just as visible. On 4 Nov, Databricks partnership activity included Wood in energy, Afresh for grocery AI and Magic EdTech in education. On 5 Nov, Databricks was reported to be investing US$350mn in Singapore to expand AI operations, alongside further reporting on the Wood partnership. On 7 Nov, Reuters content was added to the Databricks marketplace for AI use cases.

The developer surface also widened. Between 30 Oct and 8 Nov, Databricks created or surfaced repositories including databricks-sql-cli, dbt-databricks, databricks-sql-python, databricks-sql-nodejs, databricks-ai-bridge, unity-gateway, appkit, sdk-js and databricks-agent-skills. CLI v1.23.0 arrived on 31 Oct with fixes for GitHub API rate-limit errors and support for making claude and codex available on dedicated clusters via databricks ssh connect. CLI v1.24.0 followed on 8 Nov with a notable change around sensitive variable handling. This was a week of connective tissue, not a single headline product.

Fivetran spent the fortnight looking less like a narrow ingestion layer and more like a packaged transformation operator. On 29 Oct, it added documentation related to the dbt platform and troubleshooting for Salesforce and Zendesk, while removing dbt Cloud documentation and several PostgreSQL troubleshooting guides. On 30 Oct, it added troubleshooting pages for Coupa OAuth scopes, Cursor changelog content and long GitHub repository syncs, while removing Cursor AI video tutorials and PostgreSQL troubleshooting pages.

The code activity then reinforced the documentation shift. On 31 Oct, Fivetran released dbt_salesforce v2.4.2-a1 with configurable salesforce_naming_convention support, and added repositories including dbt_salesforce, dbt_package_automations, dbt_sap and dbt_jira. On 6 Nov, dbt_shopify_holistic_reporting v1.5.0 introduced a schema and data change requiring a full refresh after upgrade. On 8 Nov, dbt_salesforce v2.5.0-a2 included one schema or data change with no possible breaking changes, and dbt_zendesk was created.

Snowflake and Databricks were also filling connector and dbt gaps. Snowflake created snowflake-connector-nodejs on 31 Oct, snowflake-ml-python on 2 Nov and 5 Nov, homebrew-snowflake-cli on 3 Nov, snowflake-connector-net on 6 Nov and 8 Nov, and PowerPlatformConnectors on 7 Nov. Databricks released dbt-databricks v1.13.0 on 2 Nov, adding additive merging of tblproperties across config hierarchy levels and opt-in invocation telemetry for eligible commands. The settlement pattern is visible: the warehouse, lakehouse, connector and transformation layers are all trying to own more of the operational path around data work.

The fortnight also showed the cost of a larger surface. On 30 Oct, CVE-2026-119557 was published for a path traversal vulnerability in the n8n integration for Databricks, affecting versions 2.14.0 through 2.46.3 and allowing unauthenticated attackers to access arbitrary Databricks API endpoints. On 7 Nov, Databricks released databricks-sql-nodejs 3.0.0 with a breaking change requiring Node.js versions containing the CVE-2026-111497 Brotli decoder fix.

Snowflake had a separate security run. On 4 Nov, two CVEs naming Snowflake were published, including CVE-2026-114456, a 6.5 CVSS issue involving potential interference with user login attempts in multiple Snowflake client drivers using the authenticator=externalbrowser flow. On 6 Nov, CVE-2026-123859 was published with an 8.1 CVSS score, relating to write_pandas() data-loading functionality in the Snowflake Connector for Python when identifier escaping is disabled through a non-default configuration.

dbt Labs had operational incidents rather than CVEs. On 1 Nov, dbt Wizard had a major incident involving SQL execution failures lasting about 87 minutes, followed by a minor degraded-performance incident across the dbt platform lasting about 262 minutes. On 5 Nov, the Jobs list and Environment pages failed to load in a minor incident resolved after about 205 minutes, and a major Git error issue for projects connected to GitHub over SSH was resolved immediately.

None of this changes the direction of the market. It does clarify the price of the direction. As agents, connectors, CLIs, marketplaces and spreadsheets become part of the governed data plane, reliability and security move from back-office hygiene to buying criteria.