Teradata's AgentStack aims to simplify building, managing AI

Simplifying AI development, management

Despite heightened interest in AI development, the vast majority of all AI projects never make it into production.

Enterprise AgentStack could mark an important shift in how Teradata is used. It provides a path to operationalize AI agents at scale, allowing them to collaborate, reason over governed enterprise data and produce higher-order outputs like strategic plans rather than isolated query responses.

Fragmented infrastructure systems, disorganized data estates and the sheer complexity of building applications that can perform as or more intelligently than humans are among the many problems that stall AI projects. In response, some data management vendors are trying to help customers by providing capabilities that better enable them to discover and operationalize relevant data and simplify developing agents and other AI tools.

For example, Databricks recently launched Instructed Retriever, a new feature in its AI development environment aimed at improving the data discovery and retrieval processes to make AI applications more accurate. MongoDB, meanwhile, launched new vector embedding and ranking models that similarly address data retrieval to make AI applications more effective.

Teradata's Enterprise AgentStack is a feature set designed to make the vendor's platform not just a system of knowledge, but one that also enables customers to derive AI-powered insights, according to Sumeet Arora, Teradata's chief product officer.

"We see an intelligence revolution sweeping enterprises," he said. "Customers are increasingly focused on deriving outcomes from their enterprise data and knowledge. In conjunction with Teradata's analytics and customer intelligence applications, AgentStack enables agentic AI-driven outcomes for customers in a secure and private fashion, within the environment where their data and knowledge reside."

Each of the components that comprise Enterprise AgentStack serves a specific purpose.

AgentBuilder is a framework designed to make it faster and easier for customers to develop agents than it would be if they had to put the process together on their own. Included are integrated no-code and pro-code options, Teradata's data management and analytics capabilities to give agents contextual awareness, and industry-specific data models.

In addition, AgentBuilder features integrations with cloud service providers, Nvidia and large language model APIs, prebuilt agents for system monitoring, and SQL optimization, among others.

Similarly, Enterprise MCP, a Model Context Protocol server that enables agents to autonomously interact with data sources -- including both structured and unstructured data -- is intended to simplify and speed up the development process.

AgentStack, meanwhile, is a secure and scalable environment for deploying both individual agents as well as multi-agent systems across cloud, on premises and hybrid Teradata infrastructures. Lastly, AgentOps is a centralized interface for monitoring and managing agents, ensuring security and governance by enforcing an enterprise's policies and compliance checks.

By including different components that address the agentic AI lifecycle, Enterprise AgentStack appears to be a well-constructed feature set for building and managing agents, according to McKnight.

"It consolidates the entire agent lifecycle -- building, deploying and managing -- into a unified platform that runs directly where mission-critical data resides, solving key security and latency hurdles," he said. "Its most valuable components for this transition are AgentEngine, which enables secure deployment … and AgentOps, which mitigates economic and regulatory risks."

Donald Farmer, founder and principal of TreeHive Strategy, likewise noted that Enterprise AgentStack appears well-constructed. He pointed out that the suite does not include support of Agent2Agent Protocol (A2A), a framework that enables agents to interact with one another. But by enabling agents to share memory and data structures, Teradata is providing similar capabilities.

"Instead of agents just talking to each other -- which can lead to a game of 'Telephone' [and produce] errors -- they are operating in a shared workspace," Farmer said. "This actually may be more robust for enterprise data than simple chat-based A2A. … A shared state is much more efficient."

However, Farmer cautioned that while shared memory may enable more efficient interactions between agents in an enterprise's Teradata's environment, it may not enable the same openness as A2A, which facilitates agentic interaction across environments.

"Shared memory often implies a shared brain or controller," he said. "This is great for a single-vendor multi-agent system, but it’s the definition of a walled garden. For example, it's very hard for a Microsoft agent to step into a Teradata shared memory space."

Customers in risk averse industries such as banking and healthcare may prefer to prevent agents in one system from interacting with agents in another, Farmer continued. But with competitors including Salesforce and Microsoft touting agent ecosystems that enable agents to constantly interact, Teradata's approach comes with risk.

"Teradata is likely betting that its customers don't want autonomous negotiation yet," Farmer said. "However, if Teradata's agents can only talk to other Teradata agents within the same stack, they miss out on the broader AI economy. If the rest of the world moves toward an open Agent Exchange, Teradata risks becoming a very secure, very expensive vault that no one has the key to."

Looking ahead

Once Enterprise AgentStack is released, it will bring Teradata's agentic AI development capabilities in line with other vendors' development suites, according to McKnight. And with its focus on keeping agentic interactions within Teradata -- as Farmer noted -- the suite could particularly appeal to highly regulated enterprises.

"Where Teradata really has an edge is in running those agents safely in the real world -- especially across hybrid setups or fully air-gapped environments -- which is something most cloud-first players have a hard time supporting," McKnight said. "That matters a lot for regulated industries, and this is where AgentStack stands out."

Beyond the specific launch of Enterprise AgentStack, Teradata's product development plans center on providing customers with a platform prepared for agentic AI ecosystems, according to Arora.

"In the coming months, Teradata is focused on its vision of enabling the autonomous enterprise by leaning hard into the concept of an autonomous AI and knowledge platform ready for agentic workloads," he said. 

That focus is wise, according to McKnight, both for Teradata as well as competing vendors that now emphasize agentic AI development and deployment.

Specifically, he suggested that Teradata and its peers add features such as automated testing, cross-vendor orchestration standards, risk-based autonomy controls, sandboxing, versioning, rollback, semantic standards and human-in-the-loop oversight to enable customers to trust agents in production.

"Teradata and others need to move beyond enabling agent creation and treat AI agents like production software, with strong safety, governance and interoperability baked in," he said. "I would also encourage all autonomous agentic AI platform builders to [reinforce] the message that enterprise success will come from disciplined engineering, not AI novelty."

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.