At its annual SAS Innovate conference in Dallas this week, SAS is celebrating its 50th anniversary by making a case for its relevance in this very different era of computing. The company, which has built its reputation on statistical analysis and enterprise analytics, is now differentiating itself as a counterpoint to what its executives described as a chaotic and quickly evolving AI landscape.
For SAS, this moment is less about new AI models and more about what comes next: how organizations actually deploy, govern and trust AI at scale.
“Over the last year, the AI conversation has shifted pretty dramatically. It’s no longer this notion of how we actually use AI, or can we use AI?” said Jared Peterson, senior vice president of global engineering at SAS, during a press briefing. “We’ve started to shift into the question of, how can we start to put it to work? How do we scale it inside our organizations? How do we govern it? And how do we do all of that with trust in mind when the stakes are real?”
Last year’s SAS Innovate focused more on new AI capabilities. This year, the emphasis has shifted to execution, with announcements centered on governance, agentic AI, and the data and infrastructure required to operationalize both.
Governance Becomes Foundational
As a response to the governance challenges facing businesses, SAS announced AI Navigator, a new product designed to give organizations visibility into how AI is being used across the enterprise.
The new SAS AI Navigator (Image Courtesy of SAS)
The software inventories AI use cases, tracks dependencies between models and agents, and maps them to internal policies and external regulations. For example, a company deploying chatbots could govern both the models behind them and the policies shaping their behavior, ensuring regulatory compliance.
SAS says AI Navigator supports AI use from experimentation to deployment and consolidates oversight of both in-house and third-party assets into a single view. It is designed to address the growing gap between the rapid adoption of AI tools and the slower development of governance frameworks. The new product will be available later this year on Microsoft Azure Marketplace.
In a press conference, SAS executives said governance has moved beyond a regulatory requirement and is becoming a prerequisite for scaling AI, as trust becomes a central currency in how organizations evaluate and deploy these systems.
“When it comes to AI, particularly from a governance standpoint, we’ve got to think about the place of human judgement,” said Reggie Townsend, VP of AI ethics and governance. “AI governance is one of the ways we can help to preserve that human judgement.”
This emphasis on AI governance is not new for SAS, but as the company highlighted throughout the event, the need for it is more pressing than ever before. Enterprises are investing heavily in AI, but many lack the visibility and control needed to deploy it safely.
From Copilots to Agents
The company also announced updates to SAS Viya, its data and AI platform, which has been expanded to support agentic AI. New features include SAS Viya Copilot, described as a family of AI assistants embedded across the analytics life cycle. There is also the SAS Viya Model Context Protocol (MCP) Server, which the company says uses the open MCP standard to expose SAS Viya analytics and decisioning capabilities as tools for AI agents. Another new feature is the SAS Agentic AI Accelerator, a curated framework for building, governing and deploying AI agents within the platform.
Jared Peterson
In an interview with AIwire, Peterson described the new Viya features as a layered architecture. Copilots sit inside applications as the user-facing interface, where people interact through natural language. Behind that layer, one or more agents interpret requests, break them into steps and coordinate actions across systems.
Those agents rely on LLMs to understand user input and determine what to do next. From there, they query MCP servers, which act as a catalog of tools. Each tool represents a specific operation, allowing the agent to select and execute the appropriate function before returning the result to the user interface.
That architecture also allows the platform to operate without a traditional interface at all. Instead of interacting through an application, users or systems can call those agents and tools directly.
“Once you have an architectural approach like this, like we’ve put in Viya, you can now say, maybe not everybody needs to have that user interface experience,” Peterson said. “And because of those agents and those MCP servers, you could have a programmatic interaction with those things without ever coming from a rich visual user interface.”
SAS Viya Copilot (Image Courtesy of SAS)
The distinction reflects the bigger transition from assistive AI to systems that can take action, raising new questions about control and accountability. SAS executives repeatedly emphasized that autonomy has limits in enterprise settings.
“What we have seen in the last year is a clear distinction between agents for personal productivity and agentic AI designed for enterprise use,” said Udo Sglavo, VP of applied AI and modeling, in an interview with AIwire. “If I use an agent to schedule my calendar and it makes a mistake, I may miss a meeting. In an enterprise context, it’s unacceptable.”
As a result, SAS expects enterprise AI to move toward human-in-the-loop systems, where agents assist and automate workflows, but humans remain accountable for outcomes.
“I think the agentic AI movement for enterprise is clearly heading towards human-in-the-loop activities,” Sglavo said. “While we are hearing a lot about agents that can take over and cover the entire decision-making process, I don’t see that happening in an enterprise context because, at the end of the day, humans are responsible.”
Industry Agents and Real Workflows
Those limits on autonomy are starting to push enterprise AI in a different direction. Rather than general-purpose systems, SAS is focusing on domain-specific agents designed around defined business outcomes. The company announced new additions to its “industry accelerators” portfolio of AI agents, models, and model pipelines. This includes a Supply Chain Agent that automates sales and operations planning, along with expanded industry models for fraud detection, healthcare and public sector use cases.
Udo Sglavo
That focus on domain-specific systems also shapes how SAS is approaching what is often referred to as context engineering, the process of guiding how agents interpret and respond to user input. Rather than relying on open-ended prompts, SAS is designing its agents around predetermined business contexts, where the range of possible questions and actions is limited to a defined set of tasks.
Sglavo said much of this work happens behind the scenes. Instead of expecting users to phrase requests in a particular way, SAS maps those inputs to known prompts and workflows tied to specific outcomes. Because these agents are built for defined use cases like supply chain planning or fraud detection, the system can anticipate the types of questions being asked, even if it cannot predict how they will be phrased. The result is a more controlled interaction, where behavior is shaped less by prompt engineering at the user level and more by limits built into the agent itself.
That approach also extends to marketing, where SAS introduced multi-agent systems in its Customer Intelligence 360 platform. These agents handle tasks such as campaign design and execution, while operating under defined guardrails and human oversight.
Beyond Software: Digital Twins and Physical AI
While much of the conference focused on enterprise software, SAS also highlighted a trend toward physical AI. As AI moves beyond software, a growing number of companies are now investing in physical AI, applying it to real-world systems such as manufacturing, logistics and infrastructure.
A big hat for a big milestone: SAS marks 50 years at Innovate 2026 in Dallas
That includes digital twin technology, where SAS is combining simulation environments with analytics to model physical systems such as manufacturing facilities and healthcare operations. In one example, the company demonstrated a digital twin of a sterilization facility, used to simulate bottlenecks and train computer vision models using synthetic data.
While much of the current AI boom has focused on chat interfaces and software tools, Peterson said that greater long-term value may come from systems operating behind the scenes. In those settings, AI is applied to processes that produce tangible outputs, making its impact easier to quantify and, in some cases, harder for competitors to replicate.
“I think this is why you kind of hear this kind of steady drum beat underneath the surface right now about the phrase ‘physical AI,’” Peterson said. “I think what you’re going to see is that physical AI will be a way that people hedge against the risk of their business eroding in the software-as-a-service space. As soon as I can connect to something real that I can touch, then that’s harder for somebody to come after.”
The Takeaway
Across the many announcements at SAS Innovate, the company is clearly defining where it fits in the AI ecosystem. SAS is not interested in building frontier models but is concentrating on what surrounds them: data, governance, orchestration and domain-specific applications. Solving the hardest problems in AI right now revolves around making AI models and systems more usable and trustworthy for businesses. After 50 years in analytics, SAS is betting that this operational layer of the stack, not the models themselves, will define the next phase of enterprise AI.
The post SAS at 50: From Analytics Pioneer to the Governance and Orchestration Layer of AI appeared first on AIwire.
