Teradata expands Tera into agentic coworker for data

Teradata has expanded its Tera offering into what it describes as an agentic coworker for enterprise data work. The update introduces Tera Context Engine and Tera Harness.

The changes target data teams and business users working across enterprise systems, not just within Teradata environments.

Tera is designed to let users analyse data, build AI applications, manage infrastructure and automate workflows through natural language and guided execution. It is aimed at a range of roles, including business analysts, platform engineers and database administrators.

The announcement adds three elements to the product set: Tera Context Engine, Tera Harness and Agent Skills. They can be deployed together or used separately as part of Teradata's wider Autonomous Knowledge Platform.

New layers

Tera Context Engine is positioned as a context and orchestration layer spanning existing enterprise data estates. Teradata says it connects databases, data platforms, pipeline engines, catalogues, models and AI agents without moving data or requiring a single-vendor approach.

The tool is not limited to data managed by Teradata. It is intended to create a shared layer of business knowledge across structured and unstructured data, while linking that knowledge to governance, lineage and access controls.

This approach uses what Teradata calls a native context graph, which maps metadata, semantics and business meaning as relationships. The engine also reads from and writes back to existing systems of record, with the aim of improving context over time.

Another part of the Context Engine is a set of industry knowledge models. Teradata says these combine validated sector knowledge with statistical AI models so agents do not need to infer business meaning from scratch for every task.

Teradata argues this should reduce errors and improve traceability, especially in regulated sectors where auditability and provenance are significant concerns. It also says deterministic retrieval should lower inference costs by reducing reliance on repeated probabilistic reasoning.

Execution focus

Tera Harness is the execution layer in the new structure. It is designed to maintain context across workflows and coordinate the relevant tools, models, data and skills without manual routing by users.

The system applies execution patterns before model inference to reduce wasted iterations and control costs. It also embeds policy controls, approvals and guardrails in workflows before actions are carried out.

According to Teradata, Tera Harness is built on a Go-native engine and gRPC. In internal testing, the framework supported 512 concurrent agents on a single eight-vCPU virtual machine and handled 279 tool calls a minute.

Teradata says the harness also includes state management and checkpointing, allowing agent processes to pause for human approval and resume after failures. That is intended to support long-running workflows without requiring a separate orchestration layer.

Benchmark claims

Teradata disclosed benchmark results comparing Tera with other AI coding and data engineering tools. On SWE-bench Pro, using the same Opus 5 model, Tera used 73% fewer tokens than Claude Code, completed work 42% faster and cut total cost by 58%, while also achieving a higher task completion rate, according to the company.

On data-eng-bench, which Teradata described as a benchmark developed by Snowflake Labs and Bespoke Labs for data pipeline engineering, Tera delivered 53% lower cost per reliably solved task than Snowflake Cortex Code using Opus 5, based on published benchmark data, the company said.

Teradata also said Tera achieved the highest Pass3 score on data-eng-bench and tied for the top score on ADE-bench.

Services and rollout

Agent Skills form the third part of the launch. These are reusable functions for tasks in data engineering, data analysis and data science, and can be called by agents or used directly through natural language prompts.

Teradata says two categories of agents are included. Platform Agents are aimed at work such as workload tuning, compute sizing, telemetry and FinOps, while Analytics Agents are designed for tasks ranging from natural language query generation to SQL, Python and query optimisation.

Teradata is also offering AI Services around the product to help customers identify use cases, configure industry knowledge models and move projects into production. The aim is to help organisations avoid prolonged experimentation.

"Most enterprises are not starting from scratch with AI. They are dealing with tools that do not work together and a skills gap that makes those tools hard to use at scale. Tera is designed to work across that environment, putting business context, intelligent execution, and pre-built expertise into the hands of every person working with data. And equally important is what enterprises do not give up: control over their models, their data, and where everything runs. The result is AI that actually gets work done, at lower cost, with less overhead," said Sumeet Arora, Chief Product Officer at Teradata.

Tera Context Engine, Tera Harness and Agent Skills are scheduled to become available in the fourth quarter of 2026.