Volume 343, Issue 3 · 10 articles
Snowflake built an internal AI agent that diagnoses pipeline failures and opens a fix PR in under 16 minutes on average.
Snowflake says pipeline failures used to mean someone getting paged, digging through logs, writing a fix and opening a pull request, often night after night.
Its internal data team built an agent that now investigates every monitored pipeline failure.
The agent goes from incident detection to a ready-to-review pull request in under 16 minutes on average.
The post explains how the agent was built and how it changed the team's work as the pipeline count grew.
Snowflake introduces Decision Fast, a low-latency version of its Cortex AI decision model for real-time workflows.
Snowflake recently introduced Snowflake Decision in Cortex AI Functions in private preview for classification, scoring, filtering and routing tasks that return typed answers.
Customers asked for much faster decisions for interactive and latency-sensitive workflows such as live routing, real-time fraud checks and an agent's first triage step.
Snowflake Decision Fast addresses that need with low-latency structured decisions.
The post covers the use cases and how Decision Fast fits alongside the original model.
Snowflake's sales finance team cut monthly forecasting from over a week to hours using an AI context profile with Snowflake CoCo.
Snowflake says its sales finance team used to spend more than a week building its monthly forecast and now does it in a few hours.
The change came from an analytics-built profile that gives the AI standing context about the team's data, terminology and workflows.
With that profile, Snowflake CoCo reaches answers in less than half the steps, lowering the cost of each question even as usage grows.
The post explains how the CoCo profile works and why giving AI context up front makes answers more efficient.
Snowflake argues AI metrics must move past licenses and prompts to track what work AI is actually doing.
Snowflake says first-generation AI developer metrics measure access and activity, such as licenses, active users and prompts, but cannot show whether AI drives meaningful outcomes.
The post contrasts two organizations with the same number of AI interactions, one doing lookups and the other building pipelines and running multistep workflows, which most dashboards would score identically.
Snowflake proposes tracking what people use AI for and whether it changes high-value work.
It outlines a framework for measuring AI value beyond adoption figures.
Snowflake brings NVIDIA's new Kumo open models for graph-based predictive AI to Snowflake ML.
Snowflake says enterprise relational data encodes strong predictive signals in how customers, orders, products and accounts connect over time.
NVIDIA announced its Agent Knowledge Foundation and Kumo, a new family of open models, libraries and tools for predictive AI across structured data.
The post explains how Kumo graph ML can be scaled on Snowflake ML.
Snowflake positions the combination as a way to predict outcomes directly from relational data without heavy feature engineering.
Snowflake explains agentic analytics and the data foundation needed for AI agents to investigate business questions.
Snowflake describes agentic analytics as the next shift after dashboards and generative AI summaries, in which an AI agent investigates a business question by choosing data, learning from results and deciding next steps.
Using a customer churn example, the post shows how an agent might compare segments, review product usage and investigate support issues.
Snowflake argues that such agents require a governed, well-modeled data foundation to deliver trustworthy answers.
The post outlines the building blocks enterprises need for agent-driven analytics.
Snowflake launches the Snowflake AIM application in public preview as a dedicated home for database migrations.
Snowflake says database migrations suffer from fragmented tooling and zero risk tolerance, with teams stitching together extraction, conversion, deployment and validation tools.
Snowflake AIM began as a skill in the Snowflake CoCo CLI, then arrived in CoCo Desktop.
As of October 20 it has its own surface, the Snowflake AIM application, now in public preview.
The post walks through how the application streamlines planning, conversion and validation of migrations.
Snowflake benchmarks Elastic Channels for Snowpipe Streaming, reaching 41 GB/s of ingestion into a single table.
Snowflake recently made Elastic Channels for Snowpipe Streaming generally available, letting applications, devices and services send events directly while Snowflake manages ingestion scale.
To test performance, the team streamed a full ski season's worth of simulated resort sales into the system.
The benchmark reached 41 GB per second into a single table.
The post details the test setup, results and what they mean for high-volume streaming workloads.
Snowflake outlines how to build real-time fraud detection with online machine learning on its platform.
Snowflake says fraud keeps driving compliance costs higher, and real-time prevention has become one of financial services' most critical challenges.
Effective fraud detection needs ML infrastructure that continuously analyzes transaction patterns, flags anomalies and adapts to new threats.
The post explains how to catch bad actors before a transaction clears using production-grade online ML.
It walks through the architecture for a modern fraud detection platform on Snowflake.
Snowflake shares how Multnomah County and San Francisco modernized legacy data infrastructure to become AI ready.
Snowflake recaps a Snowflake Summit session where leaders from Multnomah County, Oregon and the City and County of San Francisco discussed modernizing government technology.
The post says legacy data infrastructure is no longer just an IT inconvenience as federal mandates, AI adoption and constituent expectations raise the stakes.
It covers the real costs of delaying modernization amid tight budgets and complex legacy systems.
The agencies describe how moving to a modern data platform prepared them for AI.