Timothy Spann

Posted on Mar 9

All Data and AI Weekly #232-09March2026

#mcp #llm #agenticai #snowflake

All Data and AI Weekly #232-09March2026

( AI, Data, NiFi, Iceberg, Polaris, Streamlit, Flink, Kafka, Python, Java, SQL, MCP, LLM, RAG, Cortex AI, AISQL, Search, Unstructured Data )

Tim's Corner: SNACK-AI Innovations

Welcome to issue 232! This week is packed with exciting developments across the Snowflake ecosystem. We're seeing major advances in semantic views, Cortex Code extensibility, and the OpenFlow Oracle connector going GA. The MCP Conference in NYC is just days away, and there's never been a better time to dive into building stateful AI agents. Let's explore!

Upcoming Events: Catch Tim Live!

For in-person I have a new sticker!

March 11: MCP Conference NYC - NYC Tim Speaking
March 12: Data for Breakfast NYC - NYC Hands on Labs
March 18: Data for Breakfast Princeton - Princeton
March 19: Warsaw IT Days - Virtual
March 24: NYC Snowflake Users Group - NYC Hands on Workshop
April 14: D4BNA - Virtual US
April 15: Nodes AI Conference - Virtual
April 13-17: D4BV - Virtual APAC
May 20: Budapest Data and AI - Budapest, Hungary - Tim Virtual speaking

All Snowflake Events: https://www.snowflake.com/about/events/

Special Discount Codes:

MCP Conference: 50% off with SPEAKER SOCIAL

Featured: Cortex Code Extensibility

Snowflake has released comprehensive documentation on Cortex Code Extensibility, enabling developers to extend and customize their AI-powered coding assistant workflows.

Key Capabilities:

Custom skill development and integration
Extended tool configurations
Workflow automation patterns
Integration with external systems

Semantic Views Updates

Semantic Views with Tableau Integration

The Snowflake engineering team published a deep dive on integrating Semantic Views with Tableau, enabling business users to leverage curated semantic layers directly in their visualization workflows.

Range Joins Support (Feb 25)

Semantic Views now support custom range joins, allowing more flexible temporal and interval-based queries against your semantic models.

Documentation:

Semi-Additive Metrics (Mar 5)

New support for semi-additive metrics in Semantic Views enables accurate calculations for metrics like inventory levels, account balances, and headcount that shouldn't be simply summed across time dimensions.

Cortex AI & Agents

AGENT_RUN Function

The AGENT_RUN function provides a programmatic interface for executing Cortex Agents directly from SQL, enabling seamless integration of AI agents into data pipelines and applications.

Documentation: https://docs.snowflake.com/en/sql-reference/functions/agent_run-snowflake-cortex

AI Complete Document Intelligence

Cortex AI's Document Intelligence capabilities have expanded, providing enhanced extraction and understanding of unstructured documents including PDFs, images, and complex layouts.

AI Functions Cost Management (GA - Mar 2)

Monitor and control your Cortex AI Functions spending with new account usage views and cost management features now Generally Available.

Documentation:

Using AI to Improve AI

The Snowflake engineering blog explores innovative approaches to using AI systems to enhance other AI systems, covering evaluation frameworks, automated prompt optimization, and self-improving architectures.

Agent World Model for Agentic RL

Deep dive into Agent World Models for agentic reinforcement learning - a novel approach enabling AI agents to build internal models of their environment for better decision-making.

Demystifying Evals for AI Agents

Anthropic published an excellent engineering guide on evaluation strategies for AI agents, covering metrics, benchmarks, and practical approaches to measuring agent performance.

Apache Iceberg Updates

Iceberg v3 Support (Preview)

Snowflake's Apache Iceberg v3 support is now in Preview, bringing enhanced partitioning, new data types, and improved interoperability.

Documentation:

Delta Deletion Vectors (Mar 2)

Support for Iceberg Delta deletion vectors enables efficient row-level deletes without rewriting entire data files.

Catalog-Linked Database Config (Mar 6)

New SYSTEM$GET_CATALOG_LINKED_DATABASE_CONFIG function for retrieving configuration details of catalog-linked databases.

OpenFlow: Oracle CDC Connector (GA)

The Snowflake OpenFlow Connector for Oracle is now Generally Available! This agentless CDC solution enables near real-time data replication from Oracle to Snowflake.

Key Features:

Agentless architecture - no software to install on Oracle servers
Real-time change data capture
Automatic schema evolution
Transaction consistency guarantees

Documentation:

Community Article: No Data Left Behind with Snowflake OpenFlow

Snowflake Postgres & pg_lake

Postgres Time Series with Iceberg

Engineering deep dive on building time series solutions with Snowflake Postgres and Apache Iceberg integration.

pg_lake Hits 1.4K Stars

The open source pg_lake project continues its momentum, now with 1,400+ GitHub stars! pg_lake brings Iceberg and data lake capabilities directly to PostgreSQL.

The New Stack Coverage: pg_lake Comes to Snowflake Postgres

Postgres Tools

pgSchema: Visual schema management - https://www.pgschema.com/
pgConsole: Web-based Postgres console - https://www.pgconsole.com/

Cortex Code Resources

Best Practices Guide

Comprehensive guide covering Cortex Code CLI best practices for data exploration, code generation, and workflow optimization.

CLI Reference

Complete command reference for the Cortex Code CLI.

Documentation: https://docs.snowflake.com/en/user-guide/cortex-code/cli-reference

UDF Debugging Skill

New skill for debugging User-Defined Functions with Cortex Code, making it easier to troubleshoot Snowpark functions.

GitHub: https://github.com/Snowflake-Applied-Field-Engineering/cortex-code-skill-udf-debugging

CortexChecker

Tool for validating and checking Cortex configurations and deployments.

GitHub: https://github.com/Snowflake-Applied-Field-Engineering/cortexchecker

From the Community: Medium & Blogs

Article	Description
Mastering Semantic Views and Cortex Agents with Cortex Code	Comprehensive guide to building semantic views and agents
Debugging Snowpark Functions with Cortex Code	Step-by-step debugging workflows
From Legacy BI to an Intelligence Layer	Lessons from Snowflake's own BI migration
Snowflake Atlas for Data Quality in Private Equity	Real-world data quality optimization

dbt Labs Updates

dbt Core Version Support (Mar 2)

New dbt Core versions are now supported for dbt Projects on Snowflake.

Documentation: https://docs.snowflake.com/en/user-guide/data-engineering/dbt-projects-on-snowflake-dbt-core-versions

ADE-Bench

The Analytics Data Engineering Benchmark (ADE-Bench) from dbt Labs provides standardized benchmarks for evaluating data engineering workflows.

GitHub: https://github.com/dbt-labs/ade-bench

Education: Building AI Agents on Coursera

New Coursera course: Building AI Agents with Snowflake

What You'll Learn:

Create AI agents that autonomously solve business problems
Enable agents to query databases and search documents via natural language
Evaluate agent reliability and optimize with orchestration instructions
Use Model Context Protocol (MCP) for external integrations

Course Link: https://www.coursera.org/learn/building-ai-agents-with-snowflake

MCP & Agent Tools

Cortex MCP Quick Suite

Quick-start templates for building MCP-enabled Cortex applications.

GitHub: https://github.com/sfc-gh-mmarzillo/cortex-mcp-quick-suite

Cortex MCP AgentCore

Core framework for building MCP-based agents with Cortex.

GitHub: https://github.com/sfc-gh-mmarzillo/cortex-mcp-agentcore

CortexCode Auto Agent

Automated agent workflows with Cortex Code.

GitHub: https://github.com/sarathi-aiml/CortexCode_Auto_Agent

Claude Platform Marketplace

Anthropic's Claude Platform Marketplace for discovering and sharing Claude-based solutions.

Link: https://claude.com/platform/marketplace

StreamNative Remote MCP Server

Webinar on implementing remote MCP servers with StreamNative.

Register: https://hs.streamnative.io/webinar-introducing-the-streamnative-remote-mcp-server

Developer Resources

Event Table Setup for Logging

Guide to setting up event tables for logging and tracing in Snowflake.

Documentation: https://docs.snowflake.com/en/developer-guide/logging-tracing/event-table-setting-up

Restricted Caller's Rights in Streamlit

New security feature enabling restricted caller's rights in Streamlit applications.

Documentation: https://docs.snowflake.com/en/developer-guide/streamlit/features/restricted-callers-rights

Enterprise AI

Scaling Enterprise AI Agents

How to scale AI agents from pilot to 6,000 users - lessons from enterprise deployments.

Public Sector Data-Informed Intelligence

Building data-informed Snowflake Intelligence solutions for public sector organizations.

People, Data, and AI Governance

Best practices for governance across people, data, and AI systems.

Engineering Deep Dive: Jacobi Forcing

The Snowflake AI Research team published a breakthrough paper on Jacobi Forcing - a training technique that transforms autoregressive transformers into causal parallel decoders.

Key Results:

Up to 4.5x higher tokens-per-forward
4x speedup on HumanEval benchmarks
3.7x speedup on GSM8K reasoning tasks
Preserves efficient KV-cache reuse
No extra drafting-verification stack needed

Ecosystem & Open Source

Apache Parquet: Variant Type

New support for semi-structured data in Apache Parquet with the Variant type.

Apache Gluten

High-performance native execution engine for Spark workloads.

Website: https://gluten.apache.org/

Open Semantic Interchange

Open standard for semantic model interchange across platforms.

Website: https://open-semantic-interchange.org/

Polymarket API

API documentation for the Polymarket prediction market platform.

Documentation: https://docs.polymarket.com/api-reference/introduction

GitHub Trending

Repository	Description
pixeltable/pixeltable	AI data infrastructure for multimodal applications
KeygraphHQ/shannon	Information-theoretic data analysis
EpicenterHQ/epicenter	Event-driven architecture framework
ValueCell-ai/ClawX	AI-powered data extraction
egoist/sitefetch	Fast website content fetcher
BIT-DataLab/Edit-Banana	Image editing framework
giovantenne/lastsignal	Signal processing toolkit
iamsrikanthnani/pluely	Data pipeline orchestration
vustagc/pianoterm	Terminal-based piano

Security & OSINT Tools

apurvsinghgautam/dark-web-osint-tools - OSINT tools collection
josh0xA/darkdump - Dark web search utility

Additional Resources

EY Open Science Challenge: https://challenge.ey.com/
Weather Data Attribution: https://developer.apple.com/weatherkit/data-source-attribution/
Vis de Urbel: https://visdeurbel.nl/en/
ClickHouse Benchmarks: https://benchmark.clickhouse.com/

SNACK-AI: The 2026 Pattern

As we look forward in 2026, SNACK-AI is the architectural pattern for powering AI applications at any scale - from mobile phones and robots to massive enterprise clusters.

SNACK-AI integrates: