All Data and AI Weekly #238-20April2026
( AI, Data, Agentic AI, Cortex Code, NiFi, Iceberg, Polaris, Streamlit, Python, Java, SQL, MCP, LLM, RAG, Cortex AI, AISQL, Search, Unstructured Data )
Tim's Corner: Agents Get Report Cards, Polaris Graduates, and JSON Goes Turbo
Welcome to issue 238! This week Snowflake delivered a trio of engineering deep dives that show the platform maturing fast. Cortex Agent Evaluations hit GA with the research-backed Agent GPA framework - Goal, Plan, Action metrics that catch 95% of human-annotated agent errors so you can ship agents with confidence instead of crossed fingers. Apache Polaris graduated to an Apache Top-Level Project and Snowflake published a compelling engineering blog on how the Iceberg REST Catalog is ending vendor lock-in for real. Meanwhile, PARSE_JSON got a 2x speed boost via simdjson integration - 200 million queries a day just got faster without changing a line of code. On the platform side, Performance Explorer gained tabs, filter presets, CSV export and search, while Dynamic Table primary keys went GA enabling incremental refresh downstream of full-refresh tables. The AI Pulse April 2026 livestream is Tuesday April 21 - don't miss the demos on Snowflake Intelligence and Cortex Agents. I'm speaking at Nodes AI Conference this week and gearing up for Summit in June. Let's dig in.
Tim's Content
Snowflake Summit 2026 - Tim is Speaking!
I'll be speaking at Snowflake Summit 2026 on NiFi and Agents. Check the session catalog:
New SNACK-AI Repositories
| Repository | Description |
|---|---|
| SNACKAI-CoCo-SpringSnowGhost | Spring-themed ghost detection application built with Cortex Code and SNACK-AI |
| SNACKAI-CoCo-AIQueries | AI-powered query builder and analytics built with Cortex Code |
| SNACKAI-CoCo-PolyMarket | Prediction market analytics built with Cortex Code and SNACK-AI |
| SNACKAI-CoCo-VPSCalculator | VPS cost calculator built with Cortex Code |
Blog: Prediction Markets at Warp Speed - What the SNACKAI-CoCo-PolyMarket Repository Reveals
Presentation: Nodes AI Conference - Ghost Busting with Neo4j Graph Analytics in Snowflake (PDF)
Upcoming Events: Catch Tim Live!
- April 21: Snowflake AI Pulse - April 2026 - Virtual Livestream
- April 22: Snowflake Connect: Building Transformation Pipelines for AI-Ready Data - Virtual
- April 23: Snowflake Connect: Analytics - Virtual
- April 29: Tysons Corner Hands-On Lab: Automate Data Workflows with Cortex Code and Evolv - Tysons Corner, VA
- April 30: Snowflake Accelerate - Virtual
- May 20: Budapest Data and AI - Budapest, Hungary - Tim Virtual speaking
- June 2-5: Snowflake Summit 2026 - San Francisco, CA - Join me, I will be speaking on NiFi and Agents.
Snowflake Dev Day (In-Person): https://www.snowflake.com/en/dev-day/in-person/
All Snowflake Events: https://www.snowflake.com/about/events/
Snowflake Summit 2026 - San Francisco, June 2-5
The biggest event of the year is coming! Snowflake Summit 2026 returns to San Francisco June 2-5. Expect keynotes, hands-on labs, deep technical sessions, partner showcases, and announcements that will shape the future of the data and AI ecosystem. Whether you're a data engineer, data scientist, architect, or business leader - this is the event to be at.
Register now: https://www.snowflake.com/summit/
Featured: Cortex Agent Evaluations & the Agent GPA Framework
Snowflake shipped a comprehensive evaluation stack for Cortex Agents this week, anchored by the research-backed Agent GPA (Goal-Plan-Action) framework.
Cortex Agent Evaluations (GA)
Cortex Agent Evaluations are now generally available - a complete solution for validating agent behavior and performance before and after production deployment. Define a dataset of test queries with expected responses, run evaluations, and receive structured metrics. Four built-in metrics cover tool selection accuracy, tool execution accuracy, answer correctness, and logical consistency. Custom LLM judge metrics let you encode domain-specific quality standards. Every evaluation run is fully traced and integrated with Cortex Agent Monitoring for thread-level and trace-level debugging.
Read more: Cortex Agent Evaluations: Monitor, Measure and Improve Your AI Agents
Agent GPA Framework
The Agent GPA (Goal-Plan-Action) framework, developed by Snowflake AI Research and available in the open source TruLens library, evaluates agents across goals, plans, and actions. In benchmark testing on the TRAIL/GAIA dataset, Agent GPA judges caught 95% of human-annotated errors (1.8x improvement over baseline), achieved 86% error localization (vs 49% for baseline judges), and matched 570 human-annotated failures across all severity levels.
Read more: What's Your Agent's GPA? A Framework for Evaluating AI Agent Reliability
Getting Started with Cortex Agent Evaluations
Hands-on guide to running your first agent evaluation - from dataset preparation through metric analysis.
Read more: Getting Started with Cortex Agent Evaluations
Featured: Apache Polaris - The Iceberg REST Catalog and the End of Vendor Lock-In
Apache Polaris graduated to an Apache Top-Level Project on February 18, 2026, cementing its role as the vendor-neutral standard for Iceberg catalog management. Snowflake's engineering blog lays out the case clearly: Polaris implements bidirectional Iceberg REST Catalog access with no proprietary extensions, the same open source project powers Snowflake's Horizon Catalog in production, and contributions from hundreds of people across dozens of companies (Dremio, Google Cloud, Microsoft, Confluent, AWS and more) have driven explosive growth. The community has spoken - the Apache Iceberg REST catalog specification is the definitive language for data exchange.
Read more: Apache Polaris: The End of Data Vendor Lock-In
Featured: PARSE_JSON 2x Faster with simdjson
Snowflake's PARSE_JSON function now runs 2x faster on average thanks to integration with simdjson, a high-performance JSON parser leveraging SIMD instructions. With nearly 200 million queries a day using JSON parsing, this delivers meaningful performance improvements at scale - an average 5.7% latency improvement for queries using PARSE_JSON. The implementation maintains 100% backward compatibility with all Snowflake-specific JSON behaviors. No query rewrites needed. No behavior changes. Just faster execution.
Read more: Turbocharging parse_json() in Snowflake: 2x Faster Performance with simdjson
Snowflake AI Pulse - April 2026
The monthly Snowflake AI Pulse livestream returns Tuesday, April 21 at 1:00 PM PT. This month: see how business users move from questions to insights to action through a personalized agentic experience powered by Snowflake Intelligence, explore your data ecosystem with Cortex Code, and accelerate agent development with Cortex Agents. Features deep dives, live demos, and Q&A with the Snowflake Product team.
Register: Snowflake AI Pulse - April 2026
Cortex AI & Agents
Cortex Agent Skills (Preview - Apr 15)
Skills let you package tool definitions, instructions and guardrails into reusable modules that multiple Cortex Agents can share. Define a skill once, attach it to any agent.
Read more: https://docs.snowflake.com/en/release-notes/2026/other/2026-04-15-cortex-agent-skills
data_agent_run Function
New data_agent_run function for programmatically invoking Cortex Agent runs from SQL and Snowpark.
Documentation: https://docs.snowflake.com/en/sql-reference/functions/data_agent_run
MCP Connectors (Limited Access)
MCP Connectors bring Model Context Protocol support to Snowflake, enabling Cortex Agents to connect with external tools and services through the standardized MCP interface.
Documentation: https://docs.snowflake.com/en/LIMITEDACCESS/snowflake-cortex/mcp-connectors
Agentic Triage: Building AI Agents for Cybersecurity
How to build agentic triage systems that automate cybersecurity alert investigation and response using Cortex Agents.
Read more: https://quickstarts.snowflake.com/guide/agentic-triage-building-ai-agents-for-cybersecurity/
Building Cybersecurity Agents with Snowflake
End-to-end guide for building cybersecurity agents on Snowflake's platform - from data ingestion through automated threat response.
Read more: https://www.snowflake.com/en/blog/building-cybersecurity-agents-snowflake/
Agent Teams (Part 4)
Fourth installment in the series on building multi-agent teams - advanced coordination patterns for complex workflows.
Read more: https://medium.com/snowflake/agent-teams-part-4-6bae1c0bcee2
Invoice Processing Pipeline with Cortex Agents
Build an end-to-end invoice processing pipeline using Cortex Agents to extract, validate and route invoice data automatically.
Read more: https://medium.com/snowflake/building-an-invoice-processing-pipeline-with-cortex-agents-d0b5c6c3f7ba
Cortex Code Credit Manager
Skillhalla for Cortex Code - a credit management skill that tracks and controls Cortex Code credit consumption.
Resources:
Snowflake Platform Updates
Performance Explorer Usability (Apr 17)
Performance Explorer in Snowsight gets a major usability upgrade: tabbed layout with Queries, Warehouses, and Tables pages; filter presets to save and set default filter configurations; CSV export for side-panel detail tables; and keyword search within side panels.
Read more: https://docs.snowflake.com/en/release-notes/2026/other/2026-04-17-performance-explorer-usability
Dynamic Table Primary Keys (GA - Apr 16)
Primary key support in dynamic tables is now Generally Available. Snowflake can use primary keys to track row-level changes and enable incremental refresh downstream of full-refresh dynamic tables. Supports base table-defined primary keys (with RELY property), query-derived primary keys from GROUP BY and QUALIFY ROW_NUMBER() patterns, and incremental refresh on full-refresh upstream tables.
Read more: https://docs.snowflake.com/en/release-notes/2026/other/2026-04-16-dynamic-table-primary-keys
Iceberg Internal Storage
Documentation for using Iceberg tables with Snowflake-managed internal storage - simplifying Iceberg adoption without external volumes.
Documentation: https://docs.snowflake.com/en/user-guide/tables-iceberg-storage
Create Table Constraint
Updated documentation for CREATE TABLE constraints including primary key, unique, foreign key and not null constraints.
Documentation: https://docs.snowflake.com/en/sql-reference/sql/create-table-constraint
Apache Iceberg & Polaris
From Iceberg to AI: The Complete Journey
Quickstart guide walking through the full journey from Apache Iceberg tables to AI-powered analytics on Snowflake.
Read more: https://quickstarts.snowflake.com/guide/the-iceberg-to-ai-journey/
Apache Polaris Iceberg REST Catalog (Engineering Blog)
Deep engineering blog on how Apache Polaris and the Iceberg REST Catalog are ending vendor lock-in. Polaris graduated to an Apache Top-Level Project with contributions from hundreds of people across dozens of companies.
Read more: https://www.snowflake.com/en/engineering-blog/apache-polaris-iceberg-rest-catalog/
Semantic Views & Intelligence
dbt Semantic Layer vs Text-to-SQL Semantic Views
Analysis comparing the dbt Semantic Layer approach with text-to-SQL semantic views for structured data querying - tradeoffs, use cases and when to use each.
Read more: https://stkbailey.substack.com/p/dbt-semantic-layer-vs-text-to-sql
Rill: Metrics SQL and Semantic Layer
Rill brings a metrics-first SQL approach to semantic layers - define metrics once, query everywhere.
Read more: https://www.rilldata.com/blog/metrics-sql
Omni-to-Snowflake Semantic View Skill
Cortex Code skill for converting Omni semantic models to Snowflake Semantic Views automatically.
AI Models & Research
Claude Opus 4 and Claude 7 Sonnet
Anthropic releases Claude Opus 4 and Claude 7 Sonnet (also known as Claude Sonnet 4) - next-generation models with improved reasoning, code generation, and agentic capabilities.
Read more: https://www.anthropic.com/news/claude-opus-4-and-claude-sonnet-4
Gemma 4 E2B
Google releases Gemma 4 E2B - the latest in the Gemma family of open models with enhanced efficiency and multimodal capabilities.
Read more: https://blog.google/technology/developers/gemma-4/
Hermes Agent
NousResearch releases Hermes Agent - an open model fine-tuned specifically for agentic tool-use and multi-step reasoning.
Read more: https://nousresearch.com/hermes-agent/
simdjson
The simdjson library parses JSON at gigabytes per second using SIMD instructions. Now powering Snowflake's PARSE_JSON with a 2x performance boost.
GitHub: https://github.com/simdjson/simdjson
From the Community: Medium & Blogs
| Article | Description |
|---|---|
| Agent Teams Part 4 | Advanced multi-agent coordination patterns for complex workflows |
| Invoice Processing Pipeline with Cortex Agents | End-to-end invoice extraction and routing with Cortex Agents |
| Cortex Code Credit Manager | Track and control Cortex Code credit consumption with Skillhalla |
| From Omni to Snowflake Semantic Views | Automated Omni-to-Snowflake semantic view conversion |
| dbt Semantic Layer vs Text-to-SQL | Comparing semantic layer approaches for structured data querying |
| Prediction Markets at Warp Speed | Tim Spann on the SNACKAI-CoCo-PolyMarket architecture |
GitHub: Snowflake Labs & Community
| Repository | Description |
|---|---|
| cortex-code-skillhalla | Skillhalla - collection of skills for Cortex Code including credit manager |
| SNACKAI-CoCo-SpringSnowGhost | Spring ghost detection with Cortex Code and SNACK-AI |
| SNACKAI-CoCo-AIQueries | AI query builder with Cortex Code |
| SNACKAI-CoCo-PolyMarket | Prediction market analytics with Cortex Code |
| SNACKAI-CoCo-VPSCalculator | VPS cost calculator with Cortex Code |
| simdjson | Parsing gigabytes of JSON per second - now powering Snowflake's PARSE_JSON |
Developer Tools & Open Source
| Tool | Description |
|---|---|
| Lightpanda | Lightweight headless browser for web automation and scraping |
| Magika | AI-powered file type detection by Google |
| XyOps | Operations toolkit for infrastructure management |
| OpenSourcePOS | Open source point of sale system |
| Frosty | Snowflake Labs guide for building an LLM chatbot on Streamlit |
| Libretto | AI prompt management and evaluation platform |
| Rill Metrics SQL | Metrics-first SQL semantic layer |
MCP & Agent Tooling
MCP Connectors in Snowflake (Limited Access)
Snowflake introduces MCP Connectors for Model Context Protocol integration, enabling Cortex Agents to connect with external tools through standardized interfaces.
Documentation: https://docs.snowflake.com/en/LIMITEDACCESS/snowflake-cortex/mcp-connectors
Agentic Triage for Cybersecurity
Quickstart guide for building AI agents that automate cybersecurity alert triage and investigation.
Quickstart: https://quickstarts.snowflake.com/guide/agentic-triage-building-ai-agents-for-cybersecurity/
ROI of Generative & Agentic AI
Snowflake published extensive research demonstrating the radical ROI organizations are seeing from generative and agentic AI implementations.
Resources:
Data Sharing & Open Table Formats
Snowflake's vision for data sharing with open table formats enables seamless data exchange while maintaining interoperability across engines and platforms.
Read more: https://www.snowflake.com/en/blog/data-sharing-open-table-formats/
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:
- Snowflake Platform
- NiFi (Apache)
- Apache Iceberg
- Cortex AI (Agents/Search/MCP)
- Kafka (Apache Kafka)
Deep Dive into SNACK-AI:
- Read: SNACK-AI: The 2026 Pattern (Medium)
- Code: Official SNACK-AI GitHub Repository
- Weekly Context: SNACK-AI Overview
Thanks
https://github.com/timothyspann
Generated with Cortex Code for All Data and AI Weekly - April 20, 2026
![Defluffer - reduce token usage 📉 by 45% using this one simple trick! [Earthday challenge]](https://media2.dev.to/dynamic/image/width=1000,height=420,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fiekbgepcutl4jse0sfs0.png)
