Snowflake (NYSE: SNOW) has announced a suite of new AI capabilities, including Semantic View Autopilot, advancements in machine learning, and enhanced AI cost governance, designed to help enterprises achieve tangible business outcomes by ensuring AI systems are trustworthy, governed, and scalable within its AI Data Cloud.
Snowflake’s Christian Kleinerman, EVP of Product, stated that AI is rapidly becoming integral to enterprise operations. He emphasized the company’s objective to realize this future by ensuring AI agents operate on consistent business logic, behave as anticipated, and scale predictably. Kleinerman highlighted that unifying trust, governance, and execution on a single platform is key to delivering effective AI in customer environments.
A significant innovation is Semantic View Autopilot, now generally available. This AI-powered service automates the creation and governance of semantic views, providing AI agents with a shared and consistent understanding of business metrics. This aims to produce trustworthy outcomes and address the bottleneck posed by fragmented, manually defined semantic layers, which often lead to unreliable AI outputs and eroded trust. The service is designed to automatically build, optimize, and maintain governed semantic views, potentially eliminating the need for manual semantic modeling. It also builds on initiatives like the Open Semantic Interchange (OSI), establishing an interoperable semantic layer across various ecosystem leaders.
Semantic View Autopilot is engineered to learn from user activity and leverage AI-powered generation to ensure business logic remains accurate and current across Snowflake data and consumption tools, including dbt Labs, Google Cloud’s Looker, Sigma, and ThoughtSpot (with ThoughtSpot integration generally available soon). Customers can create semantic views using business definitions from Snowflake and their existing business intelligence tools. This approach is intended to minimize AI hallucinations and reduce semantic model creation time, accelerating time-to-market. Leading organizations such as eSentire, HiBob, Simon AI, and VTS are already utilizing Semantic View Autopilot to shorten data-to-insight timelines and empower data teams to focus on higher-value AI innovation. Matt Walker, CTO at Simon AI, noted that Semantic View Autopilot provides their AI systems with a consistent, governed understanding of business metrics, enabling reliable personalization and AI-driven engagement for their customers.
To accelerate the delivery of robust machine learning (ML) models, Snowflake has introduced advancements to Snowflake Notebooks, now generally available. This fully-managed, Jupyter-powered notebook is designed for end-to-end data science and ML development directly on Snowflake data. It integrates with Cortex Code in Snowsight (generally available soon), a data-native AI coding agent that automates and streamlines enterprise development, allowing users to build and deploy ML pipelines using natural language prompts. Experiment Tracking, also generally available, simplifies the comparison of training runs, sharing of results, and reproduction of high-performing models within Snowflake Notebooks, fostering a collaborative and repeatable experimentation process.
For real-time use cases, Snowflake now supports Online Feature Store and Online Model Inference, both generally available. These capabilities enable features to be served in milliseconds and predictions to be delivered at scale. By conducting training, serving, and monitoring within the Snowflake platform, organizations can operationalize ML while maintaining consistent governance from data inception to model deployment and insight generation. Aimpoint Digital is leveraging Snowflake Notebooks for ML projects, addressing use cases such as personalization, fraud detection, and predictive analytics.
Ensuring trust and reliability for mission-critical enterprise decisions powered by AI, Snowflake is introducing Cortex Agent Evaluations, which will be generally available soon. This tool helps teams deploy AI agents confidently by providing traceability, measurability, and auditability of their behavior in production. Cortex Agent Evaluations offer developers detailed insight into how agents reason, act, and respond, allowing for systematic assessment of answer correctness, tool utilization, and logical consistency. This visibility enables teams to identify errors, refine decision logic, and validate agent behavior before it impacts business operations, while also improving efficiency by preventing operational waste like redundant tool calls and escalating compute costs. WHOOP is already using Cortex Agent Evaluations within Snowflake to enhance agent quality without moving data or integrating external monitoring tools.
Snowflake is also addressing the economic sustainability of AI for enterprises through expanded cost governance capabilities in Cortex AI Functions, now generally available. This allows organizations to plan, control, and audit their AI usage with precision. Before running AI workloads, teams can proactively estimate consumption using the AI_COUNT_TOKENS function, facilitating a clear understanding of how prompt design and context size correlate with actual costs.
