Getting AI apps to production means bringing the right data to AI, regardless of where it resides within the company.
That’s easier said than done.
High-quality data isn’t born. It’s made. And it’s the making, whether that involves curating, deploying, searching, and collaborating on high-quality datasets, that continues to drag out launches of AI-based solutions.
Enter Agentic AI. This new approach to composing AI solutions is transforming the way companies conduct business, automating processes that previously required a high level of human intervention.
Why not take the same approach with data?
The Starburst AI agent makes it easier to document, query, and collaborate on data products. That drives down the time required to derive business value from new AI solutions.
In this article, I’ll look at the basics of Agentic AI and see how the Starburst AI Agent leverages this new approach to accelerate data velocity.
What is Agentic AI?
Agentic AI is a generative AI (GenAI) system that can perform complex tasks with little to no human supervision.
To understand what makes Agentic AI unique, it helps to contrast it to a standard GenAI application.
The traditional GenAI model
Simple GenAI solutions prompt a Large Language Model (LLM) with an instruction and accompanying context. The LLM parses the request and then generates the best response it can.
In other words, traditional GenAI is a simple request/response model. It’s a reactive, fixed, and generalized approach to AI.
This model works fine for simple problems that require interactivity. However, it tends to break down when faced with complex, multi-step tasks.
Most importantly, traditional GenAI is non-self-governing. Humans must take action and make decisions based on the output.
The Agentic AI model
By contrast, Agentic AI differs in several regards.
Agentic AI is:
Proactive
It can detect problems before they occur, moving to identify patterns and take action.
Adaptable
Agentic AI systems can adapt quickly to changes in their environment. For example, a website going down, or a change in market conditions.
Specialized and collaborative
An Agentic AI system can be a single agent. Alternatively, it can employ a multi-agent approach, where each agent specializes in one task and performs it effectively. A multi-agent Agentic AI solution breaks down complex problems into discrete tasks and relies on the best agent to fulfill each component of the request.
Autonomous
Agentic AI continuously assesses information and learns from past decision-making. This enables it to make decisions – rebooting a server, executing a stock trade – without human intervention.
Unlocking the agentic workforce
Agentic AI architectures achieve this by incorporating additional intelligence into their solutions. Common components include:
- Perception: Identifies information in its environment (e.g., a series of stock trades and stock value changes) and extracts relevant details.
- Memory: Stores short-term and long-term information, including results of past decision-making, to inform future decisions.
- Planning: Break down a complex task into subtasks and find the most appropriate and reliable agents to handle each task based on past interactions.
- Aggregation: Combine results from other agents into a single, final answer.
- Reflection: Uses memory to assess the accuracy of a given response and re-run a subtask if an answer doesn’t meet its quality bar.
- Tools: External systems that can perform basic tasks on behalf of the agent – e.g., calculators, calendars, etc.
Benefits of Agentic AI
Agentic AI surpasses single-model intelligence by integrating specialized agents, reasoning loops, and tool use into coordinated systems. This approach creates solutions that improve over time, adapt to business needs, and can be reused across projects. The result is not just more brilliant AI, but AI that delivers tangible, repeatable value for the enterprise.
This leads to several key benefits, including:
Accuracy
Purpose-built agents and reflection algorithms give more accurate answers over time than general-purpose LLMs do on their own.
Efficiency
Because it’s proactive and autonomous, Agentic AI can minimize the human labor needed to fulfill specific requests. That frees up people to employ their intelligence on the tasks that absolutely require it.
Scalability
Multi-agent Agentic AI systems can scale the components of the solution individually, leading to improved performance and better cost control.
Reusability
The agents, tools, and other components of an Agentic AI solution can be leveraged in other projects. This encourages the construction of new solutions from pre-built, well-tested components. That enables shipping higher-quality solutions in fewer cycles.
What is the Starburst AI Agent?
At Starburst, we’re always looking for ways to make it easier for our customers to manage data and reduce the time to Return on Investment (ROI).
Take data products, for example. Using Starburst Enterprise and Agentic AI, customers can more easily package data for consumption by downstream consumers.
Data products contain more than just data. They contain crucial business context (metadata) that describes where the data comes from, what its fields mean, how it’s used, and who’s using it.
Starburst makes using Agentic AI easy with the development of the Starburst AI agent. The Starburst AI Agent empowers data stakeholders to leverage data effectively using natural language queries. Stakeholders can also leverage the Agent to better document and polish data products for general consumption, further reducing the time to business value.
Example of the Starburst AI Agent in action
Here are just a couple of ways the Starburst AI Agent makes it easier for anyone to work with data.
Documenting a data product with the Starburst AI Agent
For a data product to be discovered and utilized effectively, it must be properly documented.
Unfortunately, most aren’t. No one has the time.
Below is one example. The Commerce Customer 360 dataset is nearly ready to launch. But without documentation, no one will be able to find or use it.
The Starburst AI Agent can leverage the context it finds in your data estate to produce a first draft of documentation in seconds. That’s normally a task that would’ve taken a human hours or days.
Users can then review the AI’s work, accept it, and edit it to add additional business context. Once complete, the AI Agent can leverage this final information to document the data product in more detail, such as by providing a comprehensive description.
Searching data products with the Starburst AI Agent
Once it’s created, the data product is documented. The AI Agent can rely on its own generated documentation to help users discover and understand it.
Users can also use natural language requests to answer questions about a data product’s metadata and generate insights. For example, you might ask it: “Tell me about order trends across our market segments.”
The AI Agent responds to these requests by generating SQL code that will satisfy them. Users can run this query immediately to see the results. SQL-savvy users—such as developers, analysts, and others—can edit it and add their own unique enhancements first.
Even better, the Starburst AI Agent can return the most appropriate level of detail depending on the user. Users can choose from one of three personas, including:
- Executive: Provides data insights only
- Analyst: Provides insights, analysis, and supporting data
- Data engineer: Provides additional information about the SQL query, data quality ratings, and other technical details
If the user finds the results helpful, they can add their generated query to the list of usage examples for the data product. This helps future stakeholders, and it also allows the AI Agent to improve its response to future requests.
Using and sharing the data product
Once a data product is ready, you can share it with other users across the company. That enables other users to use and collaborate on trusted and curated data.
Starburst Enterprise groups data into clusters, a set of compute nodes that grant access to specific data sources. Once a team has finished polishing a data product on its cluster, it can publish it so that it’s discoverable by other users.
A user on another cluster then creates a subscription to a target data product. Once connected, authorized users on that cluster can use the Starburst AI Agent to examine the data and derive insights.
Better AI with AI
If you’re like everyone else in the industry, you’re hustling to get your own AI projects out the door.
And, if you’re like everyone else, the availability of high-quality, manageable data is holding you back.
The combination of AI and data products is a powerful one-two punch, delivering faster data solutions with better insight. With the Starburst AI Agent, you can ship data faster, break down traditional barriers to data access, and enable stakeholders to innovate on new ideas and new business solutions.
