LLM Observability + Evaluations provides tools to track and optimize GenAI apps. You can monitor key metrics like groundedness and relevance, which is crucial for debugging, ensuring compliance, and tuning performance.

Provisioned Throughput is a new feature that guarantees inference performance for production-grade AI applications. This helps eliminate issues with spiky latency or throttling.

Snowflake introduced several tools:

• Data Science Agent: Turns natural language prompts into working code within Snowflake Notebooks to accelerate model experimentation.
• Cortex Agent APIs: Allow developers to build complex AI-powered apps that combine Large Language Models (LLMs), search, SQL, and analytics in a single API call.
• SPCS (Snowpark Container Services) Updates: Now supports async jobs, job execution history, and model serving directly through the UI, so you can deploy and monitor models without managing Kubernetes.

Snowflake Intelligence is a new conversational interface designed for business users. It allows them to analyze both structured and unstructured data using natural language, without needing to write any SQL.

Through a new feature called Delta Direct. It converts Delta table metadata into the Iceberg format on the fly, allowing you to access Parquet files directly without data re-ingestion or duplication.

Dynamic Tables now support externally managed Iceberg tables, feature faster refreshes with latencies as low as 15 seconds, and have better built-in handling of slowly changing dimensions (SCDs).

Yes. With the new dbt Projects in Snowflake, you can now build and run dbt natively on the platform. This eliminates the need for separate orchestration tools, infrastructure, or IDEs.

Yes, Snowpipe Streaming has been significantly upgraded. It is now 3–5x cheaper, supports inline transformations, and includes a multi-language REST interface for greater flexibility.

OpenFlow is Snowflake’s native, multi-modal data integration platform. It is designed to reduce the complexity of ETL/ELT by moving any type of data (structured, unstructured, batch, streaming) across different systems and clouds with lower latency and fewer tools to manage.

The key message is that the Snowflake platform is evolving beyond just data storage and analysis into a comprehensive environment to build with data. The announcements focused on providing native tools for data engineering, AI/ML development, and application building, empowering developers to create directly on Snowflake.

Features in "private preview" are typically available to a limited set of customers for testing and feedback. "Public preview" means the feature is more broadly available for testing but may still undergo changes before general availability. "Generally available (GA)" means the feature is fully released and supported for production use. Features "in development" are planned but not yet available. As with any forward-looking statements, actual product offerings and timelines may differ.

Snowflake is making significant strides:

• Snowflake Postgres: (In development) A fully managed, enterprise-grade PostgreSQL database solution within Snowflake, bringing a popular transactional database to developers on the platform.
• Unistore Expansion to Azure: Hybrid Tables (the technology behind Unistore) will be available in Azure (private preview), allowing businesses on Azure to combine analytical and transactional data in Snowflake.
• Security Enhancements for Unistore: Hybrid Tables will now support Snowflake’s Tri-Secret Secure (TSS) (private preview) for organizations requiring customer-managed control over encryption keys.

Snowpark Container Services (SPCS), Snowflake's extensibility runtime, will soon be generally available in GCP regions. This allows customers to deploy custom solutions (data engineering pipelines, ML model training/deployment, AI apps) in any language, securely and natively within Snowflake, across any public cloud.

• A more intuitive Replication interface (generally available) in Snowsight allows code-free configuration of failover groups, object selection, and replication scheduling.
• A simplified Bulk Failover interface (in public preview soon) in Snowsight provides a point-and-click UI for faster disaster recovery initiation.
• Schema-Level Replication in Failover Groups (generally available) allows replicating only specific schemas for cost savings.
• Underlying Replication service performance enhancements (generally available) reduce overhead and latency for faster synchronization and improved RPOs.

Snapshots (in public preview soon) create point-in-time, immutable backups of Snowflake data and objects. Once created for a specified duration, they cannot be modified or deleted, even by administrators. This helps ensure regulatory compliance, enhances cyber-resilience (e.g., against ransomware), and guarantees long-term data integrity.

Enhanced AI observability features (generally available soon) will allow users to evaluate, debug, and optimize generative AI agents and apps. This includes measuring performance, comparing LLM configurations, tracing execution steps securely within Snowflake, and reporting on metrics like relevance, groundedness, and harmfulness.

Snowflake Trail provides enhanced observability across your Snowflake environment (infrastructure, pipelines, apps, AI). Core improvements (generally available) include better Event Table ingestion latency (up to 10x faster), an updated pricing model for cost-effectiveness, and Traces for SQL Statements in Event Tables for detailed performance breakdowns of stored procedures.

Yes, Enhanced Private Link support now enables secure inbound connections from Google Cloud Platform (GCP) to Snowflake (generally available) and outbound connections from Snowflake to GCP for External Access, Functions, Stages, and Volumes (generally available).

Through sensitive data monitoring and reporting (in private preview soon), Snowflake will offer automatic detection of sensitive data, along with tagging and reporting capabilities via an interactive governance UI. Additionally, synthetic data generation (generally available) allows creating altered, privacy-compliant copies of production data for analytical use without exposing original sensitive information.

Snowflake is introducing:

• New MFA Methods: Passkey support (Touch ID, FaceID, Windows Hello, security keys) and authenticator apps (generally available).
• Programmatic Access Tokens (PATs): (Generally available) A drop-in replacement for passwords, helping to eliminate static credentials, especially for BI apps not yet supporting MFA.
• Workload Identity Federation: (Generally available soon) Allows applications and services to authenticate securely without managing their own long-lived credentials.