Table of Contents

The AI Data Cloud in Focus: An In-Depth Analysis of Snowflake’s 2025 Technology Roadmap from the Toronto World Tour

Executive Summary & The Evolving Vision of the AI Data Cloud

The Snowflake World Tour 2025, with its Toronto stop serving as a key platform for major announcements, has crystallized the company’s strategic trajectory beyond its foundational data warehousing capabilities. The event showcased a deliberate and aggressive expansion into a comprehensive, unified platform engineered for data, applications, and enterprise-grade Artificial Intelligence (AI).1 The keynotes and technical sessions collectively articulated a vision for a future where the complexities of data management, AI development, and application integration are abstracted away, leaving a seamless and powerful platform for innovation. The pivotal announcements—including the unveiling of Snowflake Intelligence, the general availability of Snowflake Openflow, and significant enhancements to the Snowflake Cortex AI suite and Snowflake Horizon Catalog—are not disparate features but interconnected pillars supporting this evolved vision of the AI Data Cloud.1

A thorough analysis of the technologies presented reveals three primary strategic thrusts that define Snowflake’s current and future direction. First is the deep embedding of Intelligence as a serverless, accessible utility directly within the data layer, primarily through the expansion of the Cortex AI suite.4 This strategy aims to democratize AI, making sophisticated capabilities available to a broad spectrum of users without requiring specialized expertise. The second thrust is

Connectivity, focused on solving the end-to-end data movement challenge. Snowflake Openflow, an open and extensible data integration service, is central to this effort, designed to unify batch, streaming, structured, and unstructured data flows into and out of the platform.6 The final strategic pillar is

Control, with Snowflake Horizon Catalog positioned as the single pane of glass for governance, security, and discovery. This centralized control plane is designed to manage all data and AI assets, including those in open formats, ensuring that as the platform’s power grows, so too does the enterprise’s ability to manage it securely and compliantly.8

These strategic initiatives are creating a powerful, self-contained ecosystem. The concurrent maturation of Openflow for bidirectional data movement, Cortex AI for in-platform processing and intelligence, and Snowflake Intelligence for user interaction is a clear architectural choice. Openflow’s design for both data ingress and egress is a critical detail, moving it beyond a simple ingestion tool.6 Simultaneously, Cortex Agents and Snowflake Intelligence are engineered to take action based on data, not merely to analyze it, which necessitates interaction with external systems.11 The synthesis of these capabilities points to a sophisticated “closed-loop” system. An AI agent residing within Snowflake can analyze internal data, determine an action is needed in an external SaaS application, and then leverage Openflow as the secure, managed transport layer to execute that action. This architecture transforms Snowflake from a passive data repository into an active, intelligent operational hub, creating significant platform value and a strong competitive position against the integrated application ecosystems of hyperscale cloud providers.

Furthermore, Snowflake is strategically embracing “openness” to build a competitive moat, directly addressing historical concerns about vendor lock-in. The deep investment in first-class support for Apache Iceberg tables allows customers to use Snowflake’s powerful query engine on data stored in their own cloud object storage, effectively decoupling Snowflake’s compute from its storage.13 This move neutralizes a key differentiator often highlighted by competitors. Similarly, by building Snowflake Openflow on the widely adopted open-source standard Apache NiFi, Snowflake lowers the adoption barrier for data engineering teams already familiar with the technology and its vast ecosystem of processors.7 This strategy redefines openness not merely as the absence of proprietary formats, but as deep interoperability. The objective is to establish Snowflake as the most performant and secure processing engine for enterprise data, regardless of where it resides or its format, positioning the platform as the central hub in a modern, multi-vendor, open-format data mesh architecture.

The Intelligence Layer: The Expansion of Snowflake Cortex AI

Snowflake Cortex AI represents the core of Snowflake’s strategy to embed intelligence directly into the data cloud. It has evolved from a collection of machine learning functions into a comprehensive, fully managed suite of AI services designed to analyze unstructured data, build sophisticated data agents, and create scalable ML workflows, all within the same secure and governed environment as the underlying data.4 The announcements at the World Tour highlighted a significant expansion of this suite, focusing on agentic capabilities, advanced Text-to-SQL accuracy, and native processing of multimodal data.

Enterprise Intelligence & Agentic Capabilities

The development of agentic AI applications is a central theme, with Snowflake providing the foundational components to build, deploy, and manage intelligent agents that can reason and act upon enterprise data.

Cortex Agents Architecture

Cortex Agents are not simply large language model (LLM) wrappers; they are sophisticated orchestration engines designed to handle complex, multi-step tasks. The architecture of a Cortex Agent is built upon a four-component workflow that enables robust and reliable execution 11:

Planning: When a user submits a request, the agent first parses it to create an executable plan. This involves breaking down complex queries into smaller subtasks, disambiguating ambiguous questions by exploring different possibilities, and routing each subtask to the most appropriate tool. For example, a query asking to compare internal sales figures with recent market trends would be split into a task for Cortex Analyst to query structured sales data and another for Cortex Search to retrieve information from unstructured market reports.11
Tool Use: Once a plan is in place, the agent executes it by invoking its available tools. The primary native tools are Cortex Analyst, for generating SQL to query structured data, and Cortex Search, for performing semantic searches on unstructured data. This tool-use capability is extensible, allowing developers to integrate custom logic via stored procedures and user-defined functions (UDFs), enabling agents to interact with external systems or perform specialized calculations.11
Reflection: After each tool execution, the agent evaluates the results to determine the next step. It can decide to iterate on a task, ask the user for clarification if the results are ambiguous, or synthesize the information to generate a final, comprehensive response. This reflective step is crucial for handling complex queries accurately and maintaining compliance within Snowflake’s secure perimeter.16
Monitoring: Deployed agents can be continuously monitored and improved. Developers can track performance metrics, analyze user interactions, and collect explicit feedback (e.g., positive/negative ratings) to refine the agent’s behavior and improve its accuracy and utility over time.11

Building Agentic Applications

The “Hands-on Lab: Building Agentic Applications in Snowflake” session provided a practical look at the developer experience for creating these agents. The process involves defining an agent object that contains its configuration, orchestration instructions, and a manifest of its available tools. These tools can include Cortex Analyst semantic models, Cortex Search services, and custom UDFs or stored procedures that encapsulate specific business logic.11 This framework allows developers to construct powerful applications that bring together the full analytical power of the Snowflake platform in a conversational interface.

AI Observability

Recognizing that moving AI from pilot to production requires robust monitoring, Snowflake has introduced AI Observability, powered by TruLens. This capability is critical for building trustworthy and reliable agents and includes three key features 16:

End-to-End Evaluation: It provides systematic evaluation of agent and app performance using techniques like LLM-as-a-judge to report on metrics such as relevance, groundedness, and harmfulness. This allows for rapid iteration and refinement.16
Comparison: Users can compare evaluation runs side-by-side to assess the quality of responses across different LLM configurations, prompts, or toolsets, helping to identify the optimal configuration for production deployment.16
Tracing: It allows developers to trace every step of an agent’s execution, from the initial prompt to context retrieval, tool use, and final LLM inference. This granular visibility is essential for debugging individual interactions and optimizing for accuracy, latency, and cost.17

Bridging Business Language and Data: Cortex Analyst & Semantic Models

One of the most significant challenges in democratizing data access is accurately translating natural language business questions into precise SQL queries. LLMs alone often fail at this task because they lack the specific business context, relationships, and logic embedded within an organization’s data. Snowflake is addressing this with a sophisticated, multi-layered semantic approach.19

The Semantic Layer Deep Dive

Snowflake’s solution to the Text-to-SQL accuracy problem consists of two key, complementary components:

Cortex Analyst Semantic Models: These models, defined in YAML, serve as a knowledge base that provides the essential business context to the LLM. They bridge the gap between business terminology and the physical database schema by capturing critical details such as descriptive names for tables and columns, definitions of business-specific metrics and filters, synonyms for common terms (e.g., “revenue” for “total_sales”), explicit relationships between tables, and custom instructions for handling domain-specific query patterns.19 This rich context guides the LLM to generate more accurate and reliable SQL queries.
Native Semantic Views: Representing a significant architectural evolution, Native Semantic Views are a new schema-level object within Snowflake that allows business logic to be stored directly in the database.21 Unlike external semantic models, these views create a durable, reusable, and governable semantic layer. They define logical tables, relationships, dimensions, and metrics, providing a consistent source of business definitions that can be used not only by Cortex Analyst for natural language queries but also by BI tools and direct SQL queries.21

The introduction of Native Semantic Views is a fundamental architectural shift. Historically, semantic layers were often defined in separate files or within specific BI tools, leading to a disconnect between the business logic and the data itself, creating maintenance overhead and the risk of inconsistent metric definitions across an organization.20 By embedding the semantic view as a first-class object within the database, Snowflake centralizes this business logic at the source. This ensures that the same semantic definition is used consistently by any tool or user that accesses the data, effectively solving the problem of “metric drift.” This elevates Snowflake’s role in the enterprise data stack from a simple data warehouse to a true, authoritative “metrics store,” becoming the definitive source of truth not just for raw data, but for the business logic that gives it meaning.

Achieving >90% Accuracy

To further enhance Text-to-SQL performance, Snowflake has developed an innovative “agentic semantic model improvement” system. This system uses a team of specialized LLM agents to automatically create and refine a semantic model.19 The process involves an

Orchestrator agent that manages the workflow, a Relationships Agent that infers table joins from correct SQL examples, and a Semantic Model Editor that reviews and edits the model to align with business logic and fix errors. This automated system was shown to boost SQL accuracy by an average of 21% on the BIRD-SQL benchmark dataset compared to using an LLM without a well-defined semantic model.19 This demonstrates a sophisticated, self-improving capability that significantly reduces the manual effort required to build a high-quality semantic layer and achieve state-of-the-art accuracy.

Unlocking Unstructured Data

A major focus of the World Tour was the expansion of Cortex AI’s capabilities to handle multimodal data—documents, images, and audio—natively within the platform. This allows organizations to manage the entire lifecycle of unstructured data processing, from ingestion and extraction to analysis and application, within a single, secure, and governed environment, eliminating the need for a patchwork of specialized third-party tools.24

Document Intelligence

Snowflake’s Document AI provides a comprehensive framework for intelligent document processing. Key functions and models include:

AI_PARSE_DOCUMENT: This function extracts text and layout elements from documents with high fidelity. It offers two modes: OCR mode for quick text extraction from simpler documents and LAYOUT mode, which preserves the structural richness of complex documents, including tables, headers, and reading order, making it ideal for enterprise Retrieval-Augmented Generation (RAG) use cases.24
AI_EXTRACT: A new and powerful AISQL function designed for scalable information extraction from documents and images. It is powered by Snowflake’s proprietary multimodal model, Arctic-Extract, which processes text, layout, and image information in a single pass. This model supports 29 languages and can intelligently normalize variable data formats like dates and currencies.24
Table Extraction: A zero-shot feature within Document AI that can identify and extract structured, columnar data from tables embedded within documents, even those with complex nested headers and rows, making the data immediately ready for analysis.24

Audio Intelligence

New Cortex AI Audio capabilities bring sophisticated speech processing directly into the data warehouse, allowing audio data to be analyzed without ever leaving Snowflake’s secure perimeter.26 The core of this functionality is the

AI_TRANSCRIBE function, which provides:

Intelligent Transcription: It converts spoken language to text across 29 languages, with automatic language detection.26
Speaker Identification (Diarization): The function can automatically distinguish between different voices in a multi-speaker audio file, answering the critical question of “who said what, when.” This transforms unstructured conversations into structured, searchable data.26
Precise Timestamping: Users can specify the desired timestamp granularity, from word-level for creating interactive transcripts to speaker-level for analyzing conversation flow.27

The system is designed for enterprise scale, handling audio files up to two hours in length and 700 MB in size.26

The various components of Cortex AI are not standalone products but are deeply interconnected parts of a cohesive intelligence system. Cortex Agents are explicitly designed to use Cortex Analyst and Cortex Search as their primary tools; the agent acts as the “brain” that orchestrates the “senses” for structured and unstructured data, respectively.11 The accuracy of Cortex Analyst, in turn, is fundamentally dependent on the context provided by Semantic Models or Native Semantic Views.19 Furthermore, the output of unstructured data processing, such as from

AI_PARSE_DOCUMENT, becomes the direct input for Cortex Search, which then powers the RAG capabilities used by Cortex Agents to provide contextually rich answers.24 This creates a virtuous cycle: unstructured data is processed, indexed for semantic search, and then made available to AI agents who can combine it with structured insights from Analyst to answer complex, multi-faceted business questions. This tightly integrated architecture represents a powerful competitive advantage that is difficult for point-solution vendors to replicate.

The Evolution of AI in SQL: Cortex AISQL

Cortex AISQL represents a paradigm shift, transforming standard SQL into a powerful AI query language for multimodal data. This feature is designed to democratize AI by allowing data analysts to leverage advanced AI capabilities using the familiar SQL syntax they already know, effectively turning every analyst into an AI engineer.29

Key Functions and Syntax

Cortex AISQL embeds a suite of powerful AI operators directly into Snowflake’s SQL engine. Core functions include 29:

AI_COMPLETE: Generates text completions or descriptions for inputs, including text and images. For example, SELECT AI_COMPLETE(image_column, ‘describe this image’).
AI_FILTER: Enables semantic filtering and joining based on conceptual similarity rather than exact key matches. For instance, … JOIN solutions ON AI_FILTER(issues.description, solutions.text).
AI_CLASSIFY: Performs text and image classification into user-defined categories, supporting multi-label classification.
AI_AGG: Aggregates insights across multiple rows of text based on a user-defined prompt, overcoming context window limitations of single-row functions.
AI_TRANSCRIBE: Transcribes audio files stored in a stage directly within a SQL query.
AI_EMBED: Generates vector embeddings for text or images for use in similarity search and clustering tasks.

Performance and Cost Efficiency

A key advantage of Cortex AISQL is its deep integration with Snowflake’s native query engine. This allows for significant performance optimizations and cost savings compared to solutions that rely on external API calls to AI models. Performance improvements are reported to range from 30% to 70%, with cost savings of up to 60% when filtering or joining data across thousands of records.29 By unifying all data types and AI processing within a single platform, Cortex AISQL eliminates data silos and consolidates tooling, democratizing AI-powered analytics across the enterprise.

Table 1: The Snowflake Cortex AI Suite: A Unified Intelligence Layer

Component	Primary Function	Core Technology/Model	Target User	Example Use Case
Cortex Agents	Orchestrate complex, multi-step tasks using a suite of tools to answer questions and take action.	Planning, Tool Use, Reflection, Monitoring workflow; orchestrates other Cortex services.	App Developer, Business User (via apps)	Build a conversational app that compares internal sales data with external market trends from documents.
Cortex Analyst	Translate natural language questions into accurate SQL queries against structured data.	Text-to-SQL powered by LLMs guided by Semantic Models or Native Semantic Views.	Business Analyst, Data Analyst	Ask “What were the top 5 selling products in the EMEA region last quarter?” in natural language.
Cortex Search	Perform low-latency, high-quality semantic search on unstructured and semi-structured data.	Retrieval-Augmented Generation (RAG) and vector search on indexed documents.	App Developer, Data Scientist	Find all legal contracts that contain clauses related to “indemnification” and “liability limits.”
Document AI	Extract text, layout, tables, and structured data from documents and images.	AI_PARSE_DOCUMENT, AI_EXTRACT functions; powered by proprietary Arctic-Extract model.	Data Engineer, Data Analyst	Process thousands of PDF invoices to extract vendor name, invoice date, and line-item details into a structured table.
Cortex AI Audio	Transcribe audio files, identify different speakers, and provide precise timestamps.	AI_TRANSCRIBE function with transcription, speaker diarization, and timestamping models.	Data Analyst, Compliance Officer	Analyze customer service calls to identify speakers, transcribe the conversation, and flag keywords for compliance review.
Cortex AISQL	Embed AI operators directly into SQL for scalable, in-database analysis of multimodal data.	Suite of AI_* functions (AI_COMPLETE, AI_FILTER, etc.) integrated into the query engine.	Data Analyst, Data Engineer	Write a single SQL query to classify customer sentiment from text reviews and join it with sales data.

The Connectivity Layer: Unifying the Data Ecosystem with Snowflake Openflow

Snowflake Openflow marks a significant strategic move to provide a native, fully managed data integration service, addressing a critical gap in the company’s end-to-end platform vision.15 Announced as generally available on AWS, Openflow is designed to revolutionize data movement by unifying all data integration workloads—from traditional batch ETL to real-time streaming and multimodal data ingestion for AI—into a single, open, and extensible service.6

Architecture and Core Concepts

Openflow’s architecture is designed for flexibility, security, and enterprise-scale performance, leveraging a proven open-source foundation.

Foundation on Apache NiFi

Snowflake Openflow is built upon the robust and widely adopted open-source framework, Apache NiFi 2.0.6 This strategic choice provides several immediate advantages. It gives users access to a mature, graphical user interface (GUI) for designing data pipelines using a flow-based programming model. More importantly, it inherits NiFi’s extensive ecosystem of nearly 300 pre-built processors, which cover a vast range of tasks including data extraction, transformation, routing, and loading, dramatically accelerating the development of data integration workflows.6

Control Plane vs. Data Plane

The architecture is logically separated into two main components, which allows for a flexible and secure operational model 6:

Control Plane: This component is fully managed by Snowflake and is responsible for the overall orchestration of data processes. It includes services for managing flow configurations, overseeing runtimes, maintaining the connector catalog, and providing observability features like performance tracking and health monitoring of data flows.6
Data Plane: This is the runtime engine that executes the data integration flows. It handles the actual collection, transformation, and transmission of data. The data plane is where the processors and connectors run, performing the hands-on work of the data pipeline.6

Deployment Models

Snowflake offers two primary deployment models for the Openflow data plane, catering to different security and operational requirements 10:

Snowflake-Hosted: In this model, the data plane runs as a fully managed service within Snowpark Container Services (SPCS). This option offers maximum simplicity and ease of use, as Snowflake handles all the underlying infrastructure management.10
Bring Your Own Cloud (BYOC): This model allows the data plane to be deployed within the customer’s own Virtual Private Cloud (VPC). While the control plane remains managed by Snowflake, the data processing occurs entirely within the customer’s network boundary. This is a critical option for organizations with strict data sovereignty, compliance, or security requirements that mandate sensitive data be preprocessed locally before being loaded into Snowflake.10

Use Cases and Implementation Patterns

Openflow is positioned as a single, unified platform capable of handling the full spectrum of modern data integration needs.

Unified Data Integration

Openflow is designed to be the central service for all data movement, supporting structured and unstructured data, in both batch and streaming modes.10 Its key capability is enabling complex, bidirectional data extraction and loading. This means it can not only ingest data

into Snowflake but also serve data from Snowflake to other downstream systems, acting as a centralized data hub.6

Real-Time CDC Streaming

A critical use case highlighted is the ability to replicate Change Data Capture (CDC) streams from external Online Transaction Processing (OLTP) databases into Snowflake in near real-time.6 This allows organizations to feed their analytical and AI applications with the most current operational data without disrupting core business systems, addressing the needs of the “Real-Time Insights at Scale” session.

Powering AI Workloads

Openflow serves as the primary on-ramp for feeding Snowflake’s AI and ML pipelines. Its ability to handle continuous, high-speed ingestion of multimodal data—such as text from documents, images from sensors, or audio from call centers—is essential for populating the datasets required by Cortex AI for processing and analysis.7

Loading Patterns

To efficiently load data into Snowflake, Openflow supports several optimized patterns 6:

Snowpipe Auto-Ingest: For file-based ingestion from cloud storage like Amazon S3.
Snowpipe Streaming: For high-throughput, low-latency ingestion of streaming data, with rates of up to 10 GB/sec.
COPY INTO Command: For bulk loading of files from internal or external stages.
Snowflake Processor PutIcebergTable: For writing data directly into Apache Iceberg tables, supporting open data lakehouse architectures.

Strategic Implications

The introduction of Openflow is a significant strategic move that fundamentally changes Snowflake’s position in the enterprise data landscape.

End-to-End Platform Play

Previously, getting data into Snowflake often required customers to license and manage separate third-party ETL/ELT tools. Openflow fills this critical gap, reducing customer reliance on external vendors and consolidating more of the data lifecycle onto the Snowflake platform.15 This move positions Snowflake as a more comprehensive, one-stop solution for data management, from initial ingestion to final analysis and action.

Competitive Landscape

Openflow is Snowflake’s direct response to the evolving competitive landscape, where rivals like Databricks have been expanding their own ingestion capabilities with offerings such as Lakehouse Flow.15 By providing a powerful, native data integration service, Snowflake aims to keep more of the data pipeline within its ecosystem, offering the convenience of a unified platform to its customers.

The tight integration between Openflow and Cortex AI creates a powerful synergy that elevates the platform beyond simple data analysis. While Cortex Agents provide the “brain” for intelligent decision-making, Openflow provides the “physical nervous system” that allows these agents to interact with the broader enterprise environment. Cortex Agents are designed to use a variety of tools, and some of these tools will inevitably need to interact with external systems—for example, to update a customer record in a CRM or trigger a workflow in a marketing automation platform.11 Openflow, with its explicit design for bidirectional data movement, provides the underlying mechanism to execute these actions reliably and securely.6 This architectural pattern, where an agent triggers an Openflow dataflow as a “tool,” transforms the platform from a place where data is merely analyzed to a place where intelligent, automated actions are orchestrated and executed across the entire enterprise.

Table 2: Snowflake Openflow: Key Features and Deployment Models

Part A: Core Features
Feature	Description
Foundation	Built on the open-source Apache NiFi 2.0 framework, providing a mature and robust foundation.
Programming Model	Utilizes a GUI-based, flow-based programming model for intuitive, visual pipeline design.
Processor Ecosystem	Inherits a rich ecosystem of nearly 300 pre-built processors for a wide range of data integration tasks.
Data Modalities	Natively handles structured, unstructured, batch, and streaming data in a single service.
Data Flow Direction	Supports bidirectional data movement, including both ingress (to Snowflake) and egress (from Snowflake).
Part B: Deployment Models
Aspect	Snowflake-Hosted Model
Management	Fully managed by Snowflake, abstracting away all infrastructure concerns.
Runtime Environment	Runs within Snowpark Container Services (SPCS) in Snowflake’s cloud environment.
Network Boundary	Data processing occurs within Snowflake’s secure VPC.
Ideal Use Case	General-purpose data integration where simplicity and speed of deployment are paramount.
Cost Considerations	Billed based on SPCS compute utilization.

The Governance Layer: Centralized Control with Snowflake Horizon Catalog

As Snowflake expands its capabilities into AI and application development, the need for a robust, unified governance framework becomes paramount. Snowflake Horizon Catalog is the company’s answer to this challenge, serving as the built-in governance and discovery layer for the entire AI Data Cloud.8 It is designed to provide a single, consistent model for managing and securing all enterprise data and AI assets, ensuring that as organizations innovate, they can do so with confidence and control.

A Unified Repository for Data and AI

Horizon Catalog’s vision extends far beyond a traditional data catalog. It is engineered as a unified, global repository that encompasses not only structured, semi-structured, and unstructured data but also the full spectrum of AI and application assets, including models, Snowflake Notebooks, Streamlit apps, and data products (listings).9 The functionality of Horizon is structured around five core pillars that address the end-to-end governance lifecycle.8

Key Features in Practice

Horizon provides a comprehensive set of tools for data stewards, security administrators, and data teams to effectively govern, discover, and collaborate on content.

Governance & Compliance

To help organizations understand and manage their data estate, Horizon offers a suite of governance and compliance features 8:

Data Classification and Tagging: Horizon can automatically classify columns containing sensitive data (e.g., PII) and allows stewards to apply user-defined object tags to track data for compliance, discovery, and resource usage.
Data Quality Monitoring: It includes built-in and custom data metric functions to monitor data quality, troubleshoot issues, and configure alerts for near-real-time notifications.
Data Lineage: Horizon provides end-to-end lineage visualization in Snowsight, allowing users to trace the flow of data from source to target tables and understand dependencies between objects.

Security & Privacy

Horizon is built on a powerful security foundation and provides granular controls to protect data 8:

Access Controls: It leverages Snowflake’s robust Role-Based Access Control (RBAC) framework to control access to all objects.
Fine-Grained Policies: Data stewards can define and apply data access policies to provide column-level security (through Dynamic Data Masking and External Tokenization) and row-level security (through Row Access Policies).
Privacy-Enhancing Technologies (PETs): For highly sensitive data, Horizon supports advanced privacy features such as Aggregation and Projection Policies, which control the types of queries that can be run on shared data, and Snowflake Data Clean Rooms, which enable secure multi-party collaboration without exposing raw data.8

Discovery & Collaboration

To ensure that governed data can be easily found and utilized, Horizon provides several discovery surfaces 8:

Universal Search: A natural language search interface that allows users to find data, apps, and models both inside and outside their organization.
Internal Marketplace: A company-exclusive site where teams can discover and access curated, internal data products, complete with wiki-like descriptions and validated SQL examples.
AI-Generated Descriptions: Horizon leverages Cortex AI to intelligently generate descriptions for tables and columns, which improves the quality of metadata and enhances the discoverability of assets through Universal Search.8

AI Governance and Security

A critical function of Horizon is to provide the necessary guardrails for the safe and responsible use of AI. As organizations increasingly deploy AI models and applications, Horizon ensures that these new workloads adhere to the same stringent governance standards as traditional data workloads.5 This includes managing access to AI models through RBAC, auditing their usage to ensure compliance, and automatically enforcing data protection policies when an AI agent or model accesses a protected dataset.

One of Horizon’s most powerful capabilities is its ability to extend governance consistently to open table formats like Apache Iceberg. This means that organizations can apply the same granular access policies, masking rules, and classification tags to their Iceberg tables stored in external cloud storage as they do to their native Snowflake tables.9 This interoperability is crucial for enterprises building open data lakehouse architectures, as it ensures a unified and consistent governance posture across their entire data estate, regardless of where the data is stored.

The proliferation of generative AI in the enterprise is met with both excitement for its potential and significant apprehension regarding the associated risks of data leakage, compliance violations, and the generation of inaccurate or harmful outputs. Without a robust and deeply integrated governance framework, widespread adoption of AI at an enterprise scale is untenable. Snowflake Horizon Catalog is strategically positioned as the solution to this fundamental barrier. The “AI Governance and Security in Snowflake” session underscores this focus. Horizon provides a comprehensive suite of governance tools—including RBAC, data masking, classification, lineage, and auditing—that are not bolted on but are natively integrated with the AI workloads running in the Cortex AI suite.5 This native integration means that when a Cortex Agent accesses a dataset, it automatically and transparently inherits all the security and privacy policies applied to that data via Horizon. The security is not an afterthought; it is woven into the fabric of the platform. By providing this unified governance plane for both data and AI, Snowflake is effectively selling

trust as a core product feature. This gives security, risk, and compliance leaders the confidence they need to approve and scale AI initiatives across their organizations, representing a crucial competitive advantage in the enterprise AI market.

The No-Code Frontier: Democratizing Analytics with Snowflake Intelligence

At the World Tour, Snowflake unveiled Snowflake Intelligence, a new, no-code agentic data analytics platform designed to fundamentally change how business users interact with enterprise data.34 This announcement represents a major step in Snowflake’s mission to democratize data access, providing a unified, conversational experience that allows non-technical users to ask questions, explore data, and uncover actionable insights without writing a single line of code.12

Platform Overview

Snowflake Intelligence is envisioned as the future of business intelligence, eliminating the traditional barriers and technical overhead that separate business users from their data.35 It functions as an intelligent, conversational interface powered by data agents. Users can engage with the platform using natural language to query both structured data from database tables and unstructured information from documents, receiving synthesized, actionable insights in return.12 The platform can generate visualizations, summarize findings, and assist users in taking action based on the insights it uncovers, closing the gap between data analysis and business impact.35

Target Audience and Capabilities

Snowflake Intelligence is designed to serve a broad spectrum of users within an organization, from business teams to technical experts.

For the Business User

The primary target audience is the non-technical business user. For this group, Snowflake Intelligence provides a self-service analytics experience that removes the dependency on data teams for routine queries and reports.35 By allowing users to have a direct conversation with their data, the platform aims to break down decision-making bottlenecks and empower employees across all departments to make faster, data-driven decisions.

For the Data Scientist

Alongside Snowflake Intelligence, the company also announced the Data Science Agent. This companion tool is an agentic assistant specifically designed to boost the productivity of data scientists and ML engineers. It automates routine and often tedious ML development tasks, such as data analysis, data preparation, feature engineering, and model training.35 By breaking down complex ML workflows into distinct steps and providing verified solutions in the form of functional ML pipelines, the Data Science Agent saves technical teams hours of manual work, allowing them to focus on higher-impact initiatives.36

Underlying Technology

Snowflake Intelligence is not a standalone product but rather a sophisticated application layer built directly on top of the existing and expanding Cortex AI suite. This tight integration allows it to leverage the full power of the underlying platform.12

Powered by Cortex AI: Under the hood, Snowflake Intelligence is powered by Cortex Agents, which orchestrate the entire workflow. When a user asks a question, the agent uses Cortex Analyst (and its associated Semantic Models or Native Semantic Views) to query structured data and Cortex Search to retrieve relevant information from unstructured documents. The responses are generated by state-of-the-art LLMs from providers like Anthropic and OpenAI, running securely within the Snowflake perimeter.12
Inherited Governance: A crucial feature is that Snowflake Intelligence runs directly within an organization’s Snowflake environment. As a result, it automatically inherits all the security controls, data masking rules, and governance policies defined in the Snowflake Horizon Catalog.12 This ensures that as data access is democratized, it is done in a secure, trusted, and compliant manner.

The launch of Snowflake Intelligence can be seen as a strategic move to provide the “killer app” for the Cortex AI platform. While providing a powerful set of AI building blocks and APIs like Cortex is valuable, it requires customers to possess the technical expertise to build applications on top of them, which can slow down adoption. Snowflake Intelligence bypasses this hurdle by delivering a fully-formed, out-of-the-box, user-friendly application that any business user can begin using immediately.12 This application serves as a powerful, real-world demonstration of the underlying Cortex technologies. A business user asking a complex question that spans structured and unstructured data is seamlessly and transparently using Cortex Analyst, Native Semantic Views, and Cortex Search in concert. This “show, don’t tell” strategy effectively lowers the barrier to entry for enterprise AI, drives consumption of the underlying Cortex services, and creates a powerful flywheel effect: the more users interact with Snowflake Intelligence, the more valuable the underlying data and semantic models become, further entrenching the Snowflake platform within the core business processes of the enterprise.

Modern Data Architectures & Best Practices on Snowflake

The sessions at the World Tour also provided deep dives into the foundational technologies and best practices required to build, scale, and manage modern data architectures on the Snowflake platform. The content covered the entire data lifecycle, from data engineering and ML workloads to migration, cost management, and observability, reflecting a mature and comprehensive approach to enterprise data management.

Foundational Data Engineering

Snowflake continues to invest heavily in its data engineering capabilities, supporting a flexible, code-first approach that caters to diverse developer preferences.

Python in Snowflake

Best practices for end-to-end data engineering with Python were a key focus, centered on the capabilities of Snowpark. Snowpark is Snowflake’s developer framework that allows data engineers and data scientists to bring complex data transformations and application logic, written in Python, Java, or Scala, directly to the data within Snowflake.37 This eliminates the need to move large datasets out of the platform for processing. Best practices include using Snowpark-optimized warehouses for memory-intensive operations and leveraging the seamless, built-in Anaconda integration to access popular open-source libraries like pandas and scikit-learn securely within the Snowflake environment.37

SQL and dbt

While Python support is expanding, SQL remains a first-class language for data transformation. The sessions highlighted the tight integration of dbt Projects within Snowflake, enabling teams to build, test, and deploy scalable and reliable data transformation pipelines using a collaborative, software-engineering-centric workflow.39

Architecting for Openness: Data Lakes on Apache Iceberg

Snowflake has made a significant investment in supporting open standards, with Apache Iceberg at the forefront. This allows organizations to build flexible, open data lakehouse architectures that avoid vendor lock-in and enable multi-engine access to data.13

Technical Deep Dive

Snowflake’s integration with Iceberg is architected around two key components 14:

External Volume: This is a named, account-level object that securely connects Snowflake to a customer’s own cloud object storage (e.g., Amazon S3, Azure Blob Storage, Google Cloud Storage). This is where the Iceberg data and metadata files reside, managed by the customer.
Catalog Options: Snowflake provides two options for managing the Iceberg catalog:

Snowflake as the Iceberg Catalog: In this mode, Snowflake takes on the responsibility for all life-cycle maintenance of the table, including operations like compaction, providing full read/write access and the performance benefits of the Snowflake platform.
External Catalog Integration: This mode allows Snowflake to connect to an external catalog, such as AWS Glue or a REST-based catalog. This is ideal for interoperability, allowing other engines like Spark to concurrently access and manage the same Iceberg tables.

This architecture provides the key benefit of decoupling storage from compute. Organizations can keep their data in an open format in their own storage accounts while leveraging Snowflake’s powerful and scalable engine for querying and processing, alongside other tools of their choice.13

Scaling AI & ML Workloads

Snowflake provides a unified and integrated environment for scaling end-to-end machine learning workflows, from experimentation to production.

End-to-End ML Pipelines

The recommended best practices cover the entire ML lifecycle 42:

Data Preparation: Utilize the Snowflake Feature Store to define, manage, store, and serve ML features, with automated refreshes from batch and streaming data sources.
Model Training: Use Snowflake Notebooks, a Jupyter-like environment, for interactive development and training of large-scale models on CPU or GPU compute. For productionizing training pipelines, Snowflake ML Jobs allow Python-based ML workloads to be run remotely and automated.
Deployment and Monitoring: The Snowflake Model Registry serves as a central repository for logging and managing all ML models. From the registry, models can be deployed for scalable inference. ML Observability provides tools to monitor model performance and drift in production, while ML Lineage tracks the end-to-end flow from source data to features, datasets, and models for reproducibility and compliance.

Efficiency and Cost

To ensure that ML workloads are both performant and cost-effective, best practices include right-sizing virtual warehouses for specific tasks (e.g., larger warehouses for training, smaller ones for inference), leveraging auto-scaling and auto-suspend capabilities, and using Snowpark’s distributed APIs to scale feature engineering and training across multiple nodes efficiently.43

The Migration Blueprint: From Legacy to the AI Data Cloud

For organizations moving from legacy data warehouses, the sessions provided a clear and proven methodology to ensure a successful migration.

Strategic Approaches

Snowflake advocates for a “pragmatic lift, adjust, and shift” approach over a traditional “lift and shift” or a full redesign.46 This strategy focuses on quickly migrating the business-critical assets that have already been identified for Snowflake, with minor adjustments for optimization, rather than undertaking a lengthy and costly complete re-architecture. This allows business users to start deriving value from the platform as quickly as possible.46

Best Practices to Avoid Failure

A successful migration requires careful planning and execution. Key best practices include 47:

Planning and Assessment: Define clear objectives and scope, and conduct a thorough inventory of existing data sources, volumes, and quality issues.
Tool Selection: Choose the right tools for the migration. Snowflake is increasingly investing in its own tools, like SnowConvert AI, which uses AI to automate and accelerate the conversion of legacy code (e.g., Teradata BTEQ, Oracle SQL) and BI reports to be Snowflake-compatible.30
Testing and Validation: Conduct comprehensive testing with realistic data samples to validate migration processes and identify issues early. After the move, perform data validation to ensure completeness and accuracy.

Economic Optimization: Advanced Cost Management

Effective cost management is critical for any organization using a consumption-based platform. The sessions provided a comprehensive framework for optimizing Snowflake spend across its three main cost drivers: compute, storage, and cloud services.50

Table 3: Comprehensive Cost Management Checklist

Category	Recommended Action	Impact / Rationale
Compute	Right-size Virtual Warehouses	Avoid over-provisioning. Start with smaller warehouse sizes (e.g., X-Small) and scale up only as needed based on query performance and SLAs.
	Set Auto-Suspend to 60 Seconds	This is a critical best practice. It minimizes the time a warehouse is running and billable while idle. The 60-second minimum aligns with Snowflake’s minimum billing charge.
	Use Multi-Cluster Warehouses for Concurrency	For high-concurrency workloads like BI dashboards, enable multi-cluster warehouses to automatically scale out to handle user load, rather than using a single, oversized warehouse.
	Isolate Workloads by Warehouse	Assign separate warehouses for different workloads (e.g., ETL, BI, Data Science) to prevent resource contention and provide better cost attribution.
Storage	Manage Time Travel & Fail-Safe Retention	Review and set the DATA_RETENTION_TIME_IN_DAYS parameter appropriately. Use shorter retention for transient or staging data to reduce storage costs.
	Use Transient Tables for Non-Critical Data	For staging or temporary data, use transient tables, which do not have a Fail-Safe period and have a maximum Time Travel of one day, significantly reducing storage overhead.
	Optimize File Sizes for Ingestion	When loading data, aim for optimally sized compressed files (100-250 MB). This improves ingestion performance and avoids overhead costs associated with processing too many small files.
	Regularly Purge Unused Tables	Identify and drop unused or stale tables to reclaim storage costs associated with both active storage and their Time Travel/Fail-Safe history.
Operations	Cluster Tables by Query Predicate	For large tables, define a clustering key on columns frequently used in WHERE clauses. This enables efficient micro-partition pruning, reducing data scanned and lowering query compute costs.
	Leverage Materialized Views	For frequent, complex queries on data that does not change often, use materialized views to pre-compute and store results, accelerating performance and reducing repeat compute costs.
	Implement Query Timeouts	Set the STATEMENT_TIMEOUT_IN_SECONDS parameter on warehouses to automatically cancel runaway queries and prevent them from consuming excessive credits.
	Use Resource Monitors	Configure resource monitors to track credit consumption by warehouses and trigger actions (e.g., send notifications, suspend the warehouse) when usage exceeds defined quotas.

End-to-End Observability with Snowflake Trail

Reflecting the increasing complexity of data pipelines and applications, Snowflake introduced Snowflake Trail, a new, built-in suite of observability capabilities designed to help developers and data engineers monitor, troubleshoot, and debug their work.17

Core Components

Snowflake Trail is built on the principles of modern application performance monitoring and utilizes the OpenTelemetry standard. Its core components include 17:

Logs, Metrics, and Traces: Comprehensive telemetry signals are collected for Snowpark code, applications, and pipelines.
Event Table: This telemetry data is stored in a special type of table with a predefined schema, allowing it to be queried and analyzed using SQL.
Effortless Setup: Telemetry collection can be enabled with a single setting, eliminating the need for complex agent installation or data export tasks.

Use Cases

Snowflake Trail provides deep visibility into the performance and behavior of data workloads, enabling users to 17:

Troubleshoot Faster: Quickly diagnose and debug issues in applications and pipelines by examining detailed logs and execution traces.
Optimize Performance: Identify performance bottlenecks in Snowpark code and SQL queries.
Monitor Compute: Gain insight into resource usage and utilization.
The observability data is accessible through Snowsight’s user interface and can also be exported to preferred third-party tools like Datadog, Grafana, and Observe for integration into broader enterprise monitoring strategies.52

The strong emphasis on Python/Snowpark development, native Git integration, and the launch of a dedicated observability suite like Snowflake Trail signals a significant trend: Snowflake is providing the tools to treat data pipelines as first-class software products. Traditional data engineering has often lacked the rigor of modern software development practices like CI/CD, comprehensive testing, and deep observability. By adopting OpenTelemetry standards—the language of Application Performance Monitoring (APM) in software engineering—Snowflake is enabling data teams to natively implement DataOps and MLOps best practices.52 This indicates a strategic effort to attract and empower a new generation of data professionals with software engineering backgrounds, further embedding the platform into modern, code-first development workflows.

Industry Applications and Demonstrating ROI

The World Tour placed a strong emphasis on translating technological capabilities into tangible business value and return on investment (ROI) across key industries. The sessions for Financial Services, Retail & CPG, and the Public Sector demonstrated how Snowflake’s unified platform is being used to solve specific, high-impact challenges.

The Future of Financial Services

The financial services industry is under intense pressure to demonstrate clear returns on its AI investments.54 The sessions detailed how Snowflake’s architecture is uniquely suited to this challenge by providing a single, secure platform for both structured financial data (e.g., trades, positions, market data) and unstructured content (e.g., research reports, news feeds, earnings call transcripts). This unified approach accelerates the development of high-value use cases, including 55:

Quantitative Research and Trading: Bringing together diverse internal and third-party data sources to generate alpha and improve execution quality.
Financial Crime Detection: Integrating data from across business lines to detect fraud and meet Know Your Customer (KYC) and Anti-Money Laundering (AML) compliance requirements.
Customer 360: Creating data-driven, personalized experiences for wealth management clients by analyzing their full financial picture to inform product suitability and identify cross-sell opportunities. A recent survey revealed that for every dollar spent on AI, organizations are seeing an average return of $1.41.57

The Future of Retail and Consumer Goods

For the retail and consumer goods industry, the focus is on leveraging data and AI to build resilience and agility in response to rapidly evolving consumer behaviors and complex supply chain dynamics.58 Snowflake’s AI Data Cloud is enabling transformation across the value chain with use cases such as 58:

Hyper-Personalized Marketing: Moving beyond simple segmentation to create truly individualized customer journeys and product recommendations.
Intelligent Demand Forecasting: Integrating external data, such as real-world event data, to create more accurate and resilient forecasting models that can adapt to disruptions.
Supply Chain Optimization: Gaining end-to-end visibility into the supply chain to optimize inventory, reduce costs, and improve fulfillment.
Dynamic Pricing: Using real-time data to adjust pricing strategies based on demand, inventory levels, and competitive pressures.

The Future of the Public Sector

Public sector organizations, including federal, state, and local government agencies, are using the AI Data Cloud to modernize legacy infrastructure, break down inter-agency data silos, and deliver on their missions more effectively.61 Snowflake’s strong security and governance posture, including certifications like FedRAMP, is a critical enabler for this sector. Key use cases include 61:

Fraud, Waste, and Abuse Prevention: Integrating data across programs to identify anomalies, detect improper payments, and protect taxpayer resources.
Public Health Analytics: Centralizing diverse health data to monitor program effectiveness, improve service coordination, and enable more equitable care.
Emergency Response: Sharing data securely and efficiently across departments and agencies to coordinate responses to natural disasters and other emergencies.
Program Compliance: Providing policy makers and auditors with fast, transparent insights into program performance and its impact on citizens.

A common thread woven through the industry applications and partner sessions is the strategic shift towards a “data product” mindset.54 This represents a fundamental change from traditional, project-based data work to treating data as a reusable, valuable asset. The case studies and sessions for wealth management and other industries consistently highlighted how organizations are packaging their unique data assets into governed, discoverable, and shareable products. Snowflake’s core platform features—Secure Data Sharing, the public Snowflake Marketplace, and the new Internal Marketplace within Horizon Catalog—are the technological underpinnings of this strategy.8 This approach enables organizations to consume these data products internally to drive efficiency and consistency, or share them externally with partners and customers to create new revenue streams. This shift, facilitated by the Snowflake platform, moves data from being a cost center to a potential profit center, offering a powerful value proposition that transcends simple analytics and reporting.

Strategic Analysis and Outlook

The collective announcements and technological deep dives from the Snowflake World Tour 2025 in Toronto paint a clear picture of a company executing a bold and cohesive strategy. Snowflake is rapidly evolving from its origins as a cloud data warehouse into a multifaceted AI Data Cloud platform, a true end-to-end ecosystem for data, applications, and intelligence. This strategic expansion places Snowflake in direct competition not only with its traditional data warehouse and data lake rivals but increasingly with AI/ML platform providers, data integration specialists, and application development platforms.

The introduction of Snowflake Intelligence, built atop the increasingly powerful and interconnected Cortex AI suite, marks a pivotal moment. By delivering a no-code, agentic analytics experience, Snowflake is making a decisive move to democratize AI and data access for every user within an enterprise, significantly expanding its addressable market beyond technical teams. This is complemented by the launch of Snowflake Openflow, a native data integration service that aims to own the entire data movement lifecycle, reducing customer dependency on third-party tools and solidifying the platform’s role as the central hub of the enterprise data architecture. Underpinning this expansion is the Snowflake Horizon Catalog, a comprehensive governance layer that provides the critical trust and control necessary for enterprises to adopt these powerful new AI capabilities securely and at scale.

For technology leaders, these developments present both a significant opportunity and a call to action. The capabilities unveiled require a re-evaluation of existing data strategies and organizational structures to fully capitalize on the potential of the AI Data Cloud.

Actionable Recommendations for Technology Leaders:

Prioritize the Semantic Layer: The success of both Cortex Analyst and the broader Snowflake Intelligence platform hinges on a well-defined semantic layer. Leaders should initiate projects to create Native Semantic Views for their most critical business domains. This effort should be cross-functional, involving business analysts, data stewards, and data engineers to ensure that business logic is accurately captured and centralized within the database, creating a single source of truth for key metrics.
Evaluate Openflow for ETL/ELT Consolidation: Organizations currently relying on a fragmented landscape of third-party ETL/ELT tools should conduct a thorough evaluation of Snowflake Openflow. Its native integration, support for bidirectional data flows, and foundation on the open-source Apache NiFi standard present a compelling case for consolidating data pipelines, reducing complexity, and lowering total cost of ownership. The BYOC deployment model should be a key consideration for use cases involving sensitive data.
Establish a Center of Excellence for Agentic AI: The development of Cortex Agents and agentic applications represents a new paradigm that blends data engineering, application development, and AI/ML skills. Organizations should consider forming a dedicated Center of Excellence (CoE) or a cross-functional “tiger team” to explore and build pilot projects using Cortex Agents. This team should be tasked with developing best practices for tool creation, agent orchestration, and leveraging AI Observability to ensure a successful transition from proof-of-concept to production.
Embrace a “Data as a Product” Operating Model: The technologies showcased, particularly the Internal Marketplace within Horizon Catalog and Secure Data Sharing, are built to support a data mesh architecture and a “data as a product” strategy. Leaders should begin to identify high-value datasets and package them as governed, discoverable, and reusable data products. This shift in operating model can unlock significant value by breaking down data silos, improving data literacy, and creating new opportunities for monetization.
Modernize Governance for the AI Era: The capabilities of Snowflake Horizon Catalog should prompt a review of existing data governance policies. Security and compliance teams should be engaged to understand how features like auto-classification, tag-based masking, and fine-grained access policies can be extended to govern not just data, but also the AI models and applications that consume it. A proactive approach to AI governance will be essential for building trust and accelerating adoption across the enterprise.

In conclusion, the Snowflake World Tour 2025 has laid out an ambitious and compelling roadmap. By seamlessly integrating intelligence, connectivity, and control into a single, unified platform, Snowflake is positioning itself not just as a repository for data, but as the central nervous system for the data-driven enterprise of the future. For organizations prepared to invest in the strategic and operational shifts required to leverage these capabilities, the potential to accelerate innovation and drive measurable business impact is substantial.

Works cited

Snowflake World Tour 2025 – Toronto, accessed September 26, 2025, https://www.snowflake.com/en/world-tour/toronto/
Snowflake World Tour, accessed September 26, 2025, https://www.snowflake.com/en/world-tour/
5 Key Takeaways From Snowflake Summit, accessed September 26, 2025, https://careers.snowflake.com/us/en/blogarticle/five-key-takeaways-from-snowflake-summit
Snowflake AI and ML | Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/guides-overview-ai-features
Snowflake for AI, accessed September 26, 2025, https://www.snowflake.com/en/product/ai/
Snowflake Openflow: Simplifying Data Integration Workloads, accessed September 26, 2025, https://www.btelligent.com/en/blog/openflow-snowflake-simplify-data-integration
Snowflake Openflow Unlocks Full Data Interoperability, Accelerating Data Movement for AI Innovation, accessed September 26, 2025, https://www.snowflake.com/en/news/press-releases/snowflake-openflow-unlocks-full-data-interoperability-accelerating-data-movement-for-ai-innovation/
Snowflake Horizon Catalog | Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/snowflake-horizon
Built-In Data Governance and Discovery with Snowflake Horizon Catalog, accessed September 26, 2025, https://www.snowflake.com/en/blog/horizon-catalog-data-governance/
About Openflow – Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/data-integration/openflow/about
Cortex Agents | Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-agents
Snowflake Unveils Snowflake Intelligence: The Future of Data Agents for Enterprise AI, accessed September 26, 2025, https://www.snowflake.com/en/news/press-releases/snowflake-unveils-snowflake-intelligence-the-future-of-data-agents-for-enterprise-ai/
Managing External Data Lakes with Apache Iceberg Tables for Snowflake – Hakkoda, accessed September 26, 2025, https://hakkoda.io/resources/apache-iceberg-tables-for-snowflake/
Apache Iceberg™ tables | Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/tables-iceberg
Snowflake Openflow: The Next Frontier in Data Integration | by Pascal Pfäffle | Medium, accessed September 26, 2025, https://medium.com/@pascalpfffle/snowflake-openflow-the-next-frontier-in-data-integration-5296571609d9
Your Enterprise Data Needs an Agent – Snowflake, accessed September 26, 2025, https://www.snowflake.com/en/blog/ai-data-agents-snowflake-cortex/
Getting Started with Snowflake Trail for Observability, accessed September 26, 2025, https://quickstarts.snowflake.com/guide/getting-started-with-snowflake-trail-for-observability/index.html?index=..%2F..index
Gain End-to-End Observability with Snowflake Trail, accessed September 26, 2025, https://www.snowflake.com/en/product/use-cases/observability/
Agentic Semantic Model Improvement: Elevating Text-to-SQL …, accessed September 26, 2025, https://www.snowflake.com/en/engineering-blog/agentic-semantic-model-text-to-sql/
Cortex Analyst & Semantic Models – by Manikandan Bellan – Medium, accessed September 26, 2025, https://medium.com/@mani.bellan/cortex-analyst-semantic-models-bda036f10c67
Overview of semantic views | Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/views-semantic/overview
Snowflake Semantic Views: Real-World Insights, Best Practices, and phData’s Approach, accessed September 26, 2025, https://www.phdata.io/blog/snowflake-semantic-views-real-world-insights-best-practices-and-phdatas-approach/
Using Snowsight to create and manage semantic views – Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/views-semantic/ui
Unlock Actionable Insights from Your Most Complex Documents with …, accessed September 26, 2025, https://www.snowflake.com/en/blog/document-ai-intelligent-document-processing/
AISQL AI_PARSE_DOCUMENT – Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/snowflake-cortex/parse-document
Unlocking Audio Insights with Snowflake Cortex AI | by Kannan K R | Aug, 2025 – Medium, accessed September 26, 2025, https://krkannan.medium.com/unlocking-audio-insights-with-snowflake-cortex-ai-872b7f43489d
Cortex AI Audio | Snowflake Documentation, accessed September 26, 2025, https://docs.snowflake.com/en/user-guide/snowflake-cortex/ai-audio
The Easy Button for Context-Rich AI Agents: Snowflake Intelligence & Cortex Knowledge Extensions, accessed September 26, 2025, https://www.snowflake.com/en/blog/easy-button-context-rich-ai-agents/
Getting Started with Cortex AISQL – Snowflake Quickstarts, accessed September 26, 2025, https://quickstarts.snowflake.com/guide/getting-started-with-cortex-aisql/index.html
Snowflake Introduces Cortex AISQL and SnowConvert AI: Analytics Rebuilt for the AI Era, accessed September 26, 2025, https://www.snowflake.com/en/news/press-releases/snowflake-introduces-cortex-aisql-and-snowconvert-ai-analytics-rebuilt-for-the-ai-era/
Snowflake Cortex AISQL (including LLM functions), accessed September 26, 2025, https://docs.snowflake.com/user-guide/snowflake-cortex/aisql?lang=it%252f
Snowflake Horizon Catalog for Built-in Governance & Discovery – Atlan, accessed September 26, 2025, https://atlan.com/know/data-governance/snowflake-horizon-101/
Snowflake Horizon Catalog: Data Governance & Discovery, accessed September 26, 2025, https://www.snowflake.com/en/product/features/horizon/
Snowflake Unveils Agentic AI, Empowering 5,200+ Enterprises – Business Analytics Review, accessed September 26, 2025, https://businessanalytics.substack.com/p/snowflake-unveils-agentic-ai-empowering
Snowflake Intelligence and Data Science Agent Deliver The Next …, accessed September 26, 2025, https://www.snowflake.com/en/news/press-releases/snowflake-intelligence-and-data-science-agent-deliver-the-next-frontier-of-data-agents-for-enterprise-ai-and-ml/
News | Snowflake Announces Agentic AI Innovations | Pipeline Publishing, accessed September 26, 2025, https://www.pipelinepub.com/news/snowflake-announces-agentic-ai-innovations
Best Practices for Python Data Transformations in Snowflake, accessed September 26, 2025, https://www.snowflake.com/webinar/thought-leadership/best-practices-for-python-data-transformations-in-snowflake-2023-05-23/
The Data Engineer’s Guide to Python for Snowflake, accessed September 26, 2025, https://www.snowflake.com/resource/the-data-engineers-guide-to-python-for-snowflake/
Snowflake for Data Engineering, accessed September 26, 2025, https://www.snowflake.com/en/product/data-engineering/
What Are Apache Iceberg Tables? – Snowflake, accessed September 26, 2025, https://www.snowflake.com/en/fundamentals/apache-iceberg-tables/
Data Lake – Snowflake, accessed September 26, 2025, https://www.snowflake.com/en/product/data-engineering/data-lake/
Snowflake ML: End-to-End Machine Learning | Snowflake …, accessed September 26, 2025, https://docs.snowflake.com/en/developer-guide/snowflake-ml/overview
Best Practices For Cost-Effective Development And Deployment On Snowflake ML, accessed September 26, 2025, https://www.youtube.com/watch?v=Wr2eatgTdR0
Optimizing Snowflake Performance at Scale: Best Practices for Large-Scale Workloads | by Anuj Rawat | Medium, accessed September 26, 2025, https://medium.com/@anuj.rawat_17321/optimize-snowflake-performance-for-large-scale-workloads-best-practices-3a8a8d8dba0f
Snowflake ML Model Development, accessed September 26, 2025, https://docs.snowflake.com/en/developer-guide/snowflake-ml/modeling
Ebook: 3 Steps to Accelerate Data Warehouse Migrations to …, accessed September 26, 2025, https://www.snowflake.com/en/migrate-to-the-cloud/3-steps-accelerate-data-warehouse-migrations-snowflake/
What Is Data Migration? Types, Strategy & Best Practices, accessed September 26, 2025, https://www.snowflake.com/en/fundamentals/what-is-data-migration/
How to Migrate to Snowflake: Best Practices for Data Teams – Tredence, accessed September 26, 2025, https://www.tredence.com/blog/snowflake-migration-best-practices
Best Practices for Migrating Historical Data to Snowflake, accessed September 26, 2025, https://www.snowflake.com/en/blog/best-practices-for-migrating-historical-data-to-snowflake/
Top 8 Snowflake Cost Optimization Strategies to Reduce Cost – Ternary, accessed September 26, 2025, https://ternary.app/blog/snowflake-cost-optimization/
Snowflake Cost Optimization: Best Practices You Can Apply Today | by Vivek – Medium, accessed September 26, 2025, https://medium.com/@vivekmcm1/controlling-snowflake-costs-a-practical-guide-for-data-teams-d00569eb68df
Snowflake Trail for Observability, accessed September 26, 2025, https://www.snowflake.com/en/product/features/snowflake-trail/
Snowflake Observerability AMA, accessed September 26, 2025, https://www.snowflake.com/webinars/thought-leadership/snowflake-observability-live-ama-jan-29/
Accelerate Financial Services 2025 – Snowflake, accessed September 26, 2025, https://www.snowflake.com/accelerate/financial-services/
Accelerate Financial Services | Snowflake, accessed September 26, 2025, https://www.snowflake.com/accelerate/apac/financial-services/
AI Data Cloud for Financial Services – Snowflake, accessed September 26, 2025, https://www.snowflake.com/en/solutions/industries/financial-services/
Snowflake Research Reveals that 92% of Early Adopters See ROI From AI Investments, accessed September 26, 2025, https://www.snowflake.com/en/news/press-releases/snowflake-research-reveals-that-92-percent-of-early-adopters-see-roi-from-ai-investments/
Data Trends 2025: Retail and Consumer Goods – Snowflake, accessed September 26, 2025, https://www.snowflake.com/resource/retail-consumer-goods-data-trends-2025/
Retail & Consumer Goods Data & AI Trends in 2024 – Snowflake, accessed September 26, 2025, https://www.snowflake.com/webinar/thought-leadership/retail-consumer-goods-data-ai-trends-in-2024/
Accelerate Retail & Consumer Goods | Snowflake, accessed September 26, 2025, https://www.snowflake.com/accelerate/apac/retail/
AI Data Cloud for Government & Education – Snowflake, accessed September 26, 2025, https://www.snowflake.com/en/solutions/industries/public-sector/
Accelerate Public Sector | Snowflake, accessed September 26, 2025, https://www.snowflake.com/accelerate/public-sector/
Data-Driven Public Services: Snowflake’s Secure Solutions, accessed September 26, 2025, https://www.snowflake.com/en/solutions/industries/public-sector-uk/
Snowflake World Tour 2025 – Sigma Computing, accessed September 26, 2025, https://www.sigmacomputing.com/events/snowflake-world-tour-2025
Siemens Builds an Enterprise-Wide Data Mesh Platform to Accelerate Innovation, accessed September 26, 2025, https://www.snowflake.com/en/customers/all-customers/case-study/siemens-1/