SAS: Viya as a universal AI platform

Contents

At its Innovate customer event, SAS unveiled extensive enhancements and improvements to the Viya platform. The goal is to enable generative AI to transition faster from pilot projects to productive enterprise processes. To achieve this, the company combines three building blocks: AI assistants for analytics workflows, an agent infrastructure based on open interfaces like MCP, and modernized data management with governance, lineage, and cloud-native analytics acceleration. At the core is the new Viya Copilot, a family of AI assistants directly embedded in the platform.

Unlike generic chatbots, these assistants are not intended to run alongside work processes but to support data scientists, developers, and business users within existing analytics workflows – for example, with Python code, model pipelines, dashboard creation, search, and explanatory analyses.

According to SAS, the Copilot integrates Microsoft Foundry and is to be successively expanded to data management, model management, and AI infrastructure. Industry-specific copilots are already available for asset and liability management and clinical data analysis, among others; further functions for financial damage prevention, planning, and supply chain optimization are in development.

New infrastructure supports open standards

In parallel, SAS is introducing a new agent infrastructure for Viya. The Viya Model Context Protocol Server is planned to make many analytics, model, and decisioning functions available to external AI agents via the open MCP standard. The Agentic AI Accelerator provides code, components, interfaces, and best practices for developing, governing, and deploying custom agents. Additionally, there is the Retrieval Agent Manager, a no-code solution based on RAG, which can transform unstructured data into context-aware responses.

Agents are only as good as the data they work with

Another announcement concerns the data foundation: new or expanded features include AI-ready data management, agentic AI and copilots, and cloud-native analytics acceleration. The reasoning is straightforward: agents are only as good as the data they work with. To improve data faster and more easily, SAS focuses on bringing analytics to the data rather than constantly moving data between platforms. SpeedyStore serves as a cloud-native analytical data platform tightly integrated with Viya, and Data Accelerator is designed to execute analyses directly in large cloud data warehouses and lakehouse architectures.

Furthermore, Viya now supports embedded analytics engines like DuckDB for local analysis of open formats such as Parquet, CSV, and JSON. This addresses a classic enterprise problem: data copies increase latency, costs, and governance risks – precisely the issues that often hinder the path to a production environment.

SAS supports this with concrete figures. In a joint IDC/SAS study, 49 percent cited non-centralized or poorly optimized cloud data environments as the biggest obstacle to AI progress. And Gartner predicts that 60 percent of AI projects will be abandoned due to a lack of AI-ready data.

The market is discovering the mundane side of AI

With these announcements, SAS joins a broad trend: Generative AI is shifting from an assistant function to a controllable automation layer over business processes and data assets. For example, in March, Oracle introduced so-called agentic applications for its Fusion Cloud applications, which deploy coordinated teams of agents for ERP, HCM, supply chain, and customer experience.

SAP offers Joule Agents to leverage data in the context of business processes and automate complex workflows. In partnership with Google Cloud, Joule Agents are to be used in SAP CX Solutions, while Gemini Enterprise serves as a hub for actions across SAP and Google Cloud platforms. Data platform providers are also moving in this direction. Snowflake refers to Snowflake Intelligence and Cortex Code as a control center for the “Agentic Enterprise,” enabling the deployment of agentic functions on a unified data platform.

Agents need less magic, more operations

In the future, the crucial question will be: Who can integrate agents most securely into existing data, process, and governance landscapes? This means MCP, A2A, semantic layers, data catalogs, lineage, access controls, and observability will become the core of AI architecture.

SAS is well-positioned here. The company is considered an AI and analytics specialist for demanding data environments. This is an advantage when companies transition from AI pilot projects to productive automation. However, SAS must demonstrate that its agents and copilots are not only cleanly controlled but also deployable quickly enough, open enough for integration, and, above all, economically attractive.

(mack)