SAS Viya Adds AI Copilots & Agent Tools

Key Takeaways

• SAS Viya introduces copilots for analytics and industry-specific workflows.
• New tools enable building and governing AI agents at scale.
• Business and analytics leaders gain more oversight for production-ready AI deployments.

SAS on April 28, 2026, expanded its SAS Viya platform at SAS Innovate in Dallas with governed AI assistants, agent infrastructure and acceleration tools. According to company officials, the updates aim to help organizations move from isolated generative AI pilots to production-ready intelligence at scale.

The announcement coincides with SAS's 50th anniversary, with Microsoft, Intel and AWS as partner sponsors. 

"With SAS Viya, organizations can pair copilots and agents with human judgment, trusted data and enterprise governance, so AI doesn't just generate outputs but drives responsible, real-world decisions, said Jared Peterson, SVP, global engineering at SAS.

Table of Contents

• New SAS Viya Capabilities
• Agentic AI in Analytics: Production Realities

New SAS Viya Capabilities

The release includes three core additions:

• SAS Viya Copilot, a family of AI assistants embedded across the analytics life cycle
• SAS Viya Model Context Protocol (MCP) Server, which exposes Viya analytics to external AI agents through the open MCP standard
• SAS Agentic AI Accelerator, a framework for building and deploying AI agents within Viya

• General Q&A across core Viya applications, including data discovery, model pipeline development, model management, decision intelligence and environment management
• Code acceleration with AI-generated SAS and Python code, documentation and explanations
• Model pipeline guidance with intelligent recommendations, next steps and explainability
• Conversational dashboarding with AI-driven data enrichment, natural language dashboard creation and automated insights
• Visual investigation with AI-assisted search and AI-powered case and alert narratives
• SAS Asset and Liability Management (ALM) Copilot, which guides analysts through scenario configuration, execution and interpretation for financial risk workflows, translating natural-language inputs into governed analytic models
• SAS Health Clinical Data Discovery Copilot, which accelerates clinical data discovery and analysis by guiding 

Agentic AI in Analytics: Production Realities

The move from pilot experiment to reliable AI workflow requires deliberate architecture, not broad autonomy. 

OpenAI's internal data agent deployment offers a useful reference. Rather than rolling out a general-purpose copilot, the company built a tightly scoped tool for querying internal data and producing decision-support outputs within predefined constraints.

The lesson: agent-driven analytics deliver the most value as purpose-built workflow components. Successful agents start as specialists handling repeatable tasks with well-understood inputs, permissions and outcomes.

"An operational AI agent is not defined by intelligence," said Colleen Goepfert, executive advisor at Peak Line Advisory. "It is defined by authority."

From Dashboards to Conversational Exploration

Agentic analytics upends the enterprise data model:

• Traditional: Dashboards, scheduled reports, SQL-heavy specialist workflows
• Agentic: Conversational exploration, natural language access, dynamically evolving queries

Product managers, engineers and operators can explore data without involving data teams, surfacing insights static dashboards cannot.

Agent-mediated analysis also introduces new failure modes. Juan José López Murphy, head of data science and AI at Globant, warned that agents "may misunderstand context, forget original intent during long-running processes or pursue objectives easier to prove as done than meaningful." 

Static policy frameworks are insufficient for agentic systems. As agents become embedded in daily work, AI governance must evolve into dynamic, human-centered design built into every interaction, paired with review and clear ownership structures.