Watch this Ask the Expert session to explore how to upskill your AI agents to perform advanced data processing, analytical and decisioning tasks, with trust and governance by design.
Watch the webinar
You will learn about:
- The fundamentals of empowering AI agents.
- The capabilities SAS offers to AI agents through MCP-enabled tools.
- Use cases illustrating the value of incorporating advanced analytics into your AI agents.
The questions from the Q&A segment held at the end of the webinar are listed below and the slides from the webinar are attached.
Q&A
What other functionality will SAS be making available through MCP tools in the near future?
You saw a sampling of it with the MCP server I demonstrated in Claude. We are continuing to build out additional capabilities, tapping into the power of Viya applications across the full ALC, so that inventory will keep growing and expanding. Next month, or within the next few months, we expect to release a capability to package entire decision flows as MCP tools that you can incorporate into your agent. SAS has an application called Intelligent Decisioning, which helps you build in calls to LLMs and other analytics, as well as incorporate business rules. All of this can be packaged as a tool to make available to agents through an MCP server. This provides a more structured, governed approach to supplying tools to your agents.
What's the difference between AI Agents and Agentic AI?
When we talk about AI agents, we're referring to the definition we adopt—a task which aligns with general market definitions. We define AI agents as AI systems that can be autonomous or semi-autonomous, possessing the agency to decide, act, and perform tasks. These are artificial intelligence systems that differ from other types of AI in their ability, when given a specific goal and objective, to use available tools to make decisions and choose actions that best achieve the goal. The distinction between autonomous and semi-autonomous is important, as we do not believe only in fully autonomous agents; for high-stakes decisions, a human in the loop should remain. That's why we differentiate between degrees of agency and human involvement. That's AI agents. When it comes to agentic AI, on the other hand, I would define it as an AI paradigm or approach that enables autonomous or semi-autonomous agents to operate with trusted decisioning power, human oversight, and governance by design. I see these three as the key pillars of a successful agentic AI paradigm: trusted decisioning, human oversight, and embedded governance. Thus, agentic AI is an AI paradigm and approach, whereas AI agents are more specifically AI systems.
How can organizations effectively empower and upskill AI agents using SAS MCP servers to ensure they deliver reliable, high-quality, and actionable outcomes?
The last part—reliable, high-quality outcomes, and actionable as well—is, for us, what we have always provided as a foundation coming from SAS Viya. By packaging these capabilities as tools to empower agents, you gain trust and reliability. It’s not the LLM calling some random tool or using whatever built-in knowledge it was trained with to answer the question; it’s using trusted tools backed by a governance framework within the Viya platform. As for operationalizing this, the MCP server tapping into the Viya end-to-end environment is not yet GA, but in Q2 it will be released as part of every Viya deployment. When organizations set up their agentic ecosystem, they will simply provide the MCP server and register it as available to their agents within their environment.
If AI agents can act autonomously, how do we make sure they don't go rogue — what stops them from making decisions we didn't intend?
That's one of the most important questions in this space right now. I’m not sure if any of the other speakers would like to add something as well. To me, the short answer is guardrails, governance, and grounding. Don't think of agents as operating in a vacuum—they must operate within boundaries that you define, including what systems they can access, what actions they’re authorized to take, and what thresholds trigger a human review before anything executes. The protocol layer, the MCP, is part of what makes that possible, because it controls exactly what an agent can and cannot connect to. The deeper answer, which really contributes to trusted autonomy, is also about the quality of intelligence the agent uses. We are hearing a lot about the importance of context engineering, and that is becoming one of the most important considerations. Traditional machine learning is excellent at leveraging specific enterprise knowledge but not general knowledge. AI agents are the opposite: they excel at general knowledge but lack specific enterprise knowledge unless they are trained properly and provided with the exact context and high-quality intelligence needed. If an agent is making decisions based on poor data, biased models, or logic that cannot be explained, that’s where things go wrong. This is why the analytical foundation matters so much. As Brett mentioned earlier, what differentiates SAS tools is that these tools have already been used for decades in production by humans making decisions. Now, we are enabling agents to access these tools to make trusted decisions. That, to me, is how you ground agents in trusted governance and analytics, so you’re not just hoping they do the right thing—you can see why they did it, what they did, and course-correct when necessary.
Viya is the platform server, and this gets to connect to MCP server - correct? Question: if the org does not run on Viya, then this does not allow connection to MCP server?
Correct. The first demo that I showed in Claude was using an MCP server deployed in a Viya environment, so yes you do have to have access to a Viya environment for that. But what Jorge just showed at the code level is an MCP server you deploy locally - but still requires a connection to Viya to run it.
Is the MCP server built as part of SAS Viya deployment?
The one for interacting with the Viya applications on the back end is part of the Viya deployment (not yet released GA). The code-based one Jorge showed is available to download and run locally.
Recommended Resources
Agentic AI from SAS
Please see additional resources in the attached slide deck.
Want more tips? Be sure to subscribe to the Ask the Expert board to receive follow up Q&A, slides and recordings from other SAS Ask the Expert webinars.
