This blog post is Human-Centered Content: Written by humans for humans.

On the flight out to Tableau Conference this past April, I set out to get up to speed on Tableau’s newest platform, Tableau Next. While digging in, one term kept popping up: Agentic AI. At first, it felt like just another bit of AI jargon. But by the time I landed — and definitely by the time I’d sat through a few sessions — it was clear this was something bigger. Agentic AI wasn’t just a buzzword at TC25. The phrase showed up in nearly every keynote, demo and hallway conversation. I soon learned the truth: it’s the not-so-hidden engine driving Tableau’s next chapter.

Tableau Next

Tableau Next (formerly known as Tableau Einstein) is Tableau’s bold step forward — a reimagined analytics platform built on top of Salesforce and deeply integrated with Agentforce, Salesforce’s system for creating and deploying autonomous AI agents.

At its core, Tableau Next blends Tableau’s familiar flexibility with a new layer of intelligent, automated and personalized insight. Here are a few of the key features:

• Tableau Semantics: An AI-powered semantic layer that translates complex data into user-friendly business terms — making analytics more accessible for everyone, even those without a technical background.
• Composable Architecture: Modular, reusable data objects that let users tailor analytics solutions to their needs.
• Actionable Insights: The ability to take action directly within the platform, streamlining the path from insight to execution.
• Agentic AI Integration: Intelligent agents like Data Pro, Concierge and Inspector assist with tasks like data prep, natural language querying and data monitoring.

This shift isn’t just about new features — it represents a real pivot. Tableau is moving away from static dashboards and toward dynamic, adaptive systems that help users engage with data more intuitively.

What Is Agentic AI?

Agentic AI refers to artificial intelligence systems built to act as independent autonomous agents. These systems don’t just provide information — they actively perform tasks, make decisions and pursue defined goals on behalf of a user or another system. The more advanced variants can handle complex, multistep tasks by planning workflows, selecting the right tools and adapting as new information becomes available.

What sets Agentic AI apart from the more familiar Generative AI (like ChatGPT or image generators) is the intent behind the output. Generative AI creates content. Agentic AI solves problems. Where generative models might answer your question, an agentic system just handles the task. It weighs options, evaluates constraints and chooses the best course of action based on its objectives and environment.

This distinction matters. In business, it’s not enough to generate content or surface insights — teams need systems that can adapt, make smart decisions and act. Agentic AI brings the assessment and initiative that turns data into impact.

The Five Types of AI Agents

Stuart Russell and Peter Norvig, in the 2020 edition of Artificial Intelligence: A Modern Approach, identify five types of AI agents, each progressively increasing in complexity and capability. These categories help us understand how different agents perceive and act within their environments:

Simple Reflex Agents

These are the most basic type of AI agents. They operate by responding to current environmental inputs using simple condition-action rules (think: “if X, then do Y”). They don’t take past events or future consequences into account — they just react in the moment.

Examples:

• A thermostat turning on the heat when the temperature drops below a certain level
• An automatic door opening when motion is detected nearby

Model-Based Reflex Agents

Model-based reflex agents are a step up. Like their simpler counterparts, they use condition-action rules, but with a key difference: they maintain an internal model of the world. This lets them make more informed decisions based on both current input and past observations.

Example: A robot vacuum that remembers which parts of the room it has already cleaned, avoiding redundant work.

Goal-Based Agents

Goal-based agents are driven by internal objectives. Instead of reacting based on fixed rules, they evaluate different possible actions based on how likely they are to achieve a desired outcome. This allows for more flexibility and forward-thinking behavior.

Example: A GPS navigation system that considers traffic, road closures and distance to find the best route to your destination.

Utility-Based Agents

Utility-based agents go one step further. Like goal-based agents, they evaluate multiple possible future states — but instead of simply choosing any path to the goal, they weigh the desirability of each option using a utility function. This is especially useful when multiple goals are in play or trade-offs are involved.

Example: A recommendation engine like Netflix’s, which suggests content by balancing recent user activity, predicted enjoyment and content diversity to maximize engagement.

Learning Agents

Learning agents are the most advanced of the five. They’re capable of improving their behavior over time by learning from experience. Instead of relying solely on pre-programmed rules or models, they adapt based on feedback from their environment.

Example: Self-driving cars that use reinforcement learning to make real-time decisions — adjusting speed, changing lanes or braking — based on both simulated and real-world driving experiences.

Tableau Agent

Formerly known as Einstein Copilot for Tableau, Tableau Agent is a central feature of the Tableau Next platform. It’s designed to act as an intelligent copilot — helping users explore their data, spot trends and communicate insights clearly and efficiently.

Rather than a single, monolithic AI, Tableau Agent is made up of multiple coordinated agents, each focused on a specific task. Together, they operate as a unified system that supports users throughout the analytic workflow.

In terms of AI architecture, Tableau Agent can be best understood as a combination of:

• Goal-Based Agent: When you ask Tableau Agent a question — like, “Show me monthly profit margin over the last six months” — it works to interpret your intent and generate the appropriate visualization. It’s not just responding to input. It’s taking purposeful action to help you reach a goal.
• Utility-Based Agent: Tableau Agent does more than just follow instructions. It considers factors like context, data visualization best practices and clarity of communication. It evaluates multiple options and selects the one that provides the most useful and relevant outcome for the user.
• Learning Agent (Emerging): While Tableau Agent doesn’t currently learn in a fully autonomous way, it’s part of Salesforce’s larger AI ecosystem — which is built to incorporate feedback loops and improve over time. As the platform evolves, we can expect Tableau Agent to become more adaptive and personalized in how it supports decision-making.

In Conclusion: The Future of Analytics

Heading into TC25, I assumed Agentic AI was just the latest buzzword in data analytics — but by the end, it was clear it’s the engine driving the next chapter. Tableau Agent, as a core underpinning of Tableau Next, shows how AI is shifting from something we query to something that works alongside us — turning questions into informed action.