Teradata Corp. is expanding its Tera artificial intelligence assistant with a context engine and an execution system that let agents carry out data tasks across enterprise systems. Tera is an agentic AI workspace and a natural-language interface: users work with enterprise data and AI agents by describing what they need instead of writing SQL queries or code. The update targets business analysts, data engineers and database administrators, and it matters because the assistant is being positioned to act on data outside Teradata's own systems while staying inside an organization's access rules.
What the Tera update includes
The release adds three components. The Tera Context Engine connects information from databases, catalogs, pipelines and other sources without requiring companies to move their data. It brings together metadata, data lineage, business definitions and access policies so an agent can interpret a request in the context of the organization using it, and Teradata says the engine can trace AI outputs back to their sources and apply the same policies as information moves between systems. Tera Harness handles execution: it selects the tools, models and data needed for a task, tracks progress across multiple steps and can pause for human approval before sensitive actions. Agent Skills packages common data engineering, analysis and data science tasks into reusable functions. Teradata says the new capabilities, together with a set of specialized agent skills, will be available in the fourth quarter of 2026.
Although Tera is part of the core Teradata Autonomous Knowledge Platform, the company says it can reach data held outside Teradata systems. That is the point of separating context from execution. The Context Engine supplies the organizational meaning of a request — which definitions apply, where the data came from, who is allowed to see it — while Harness decides what to run and in what order. Controls are applied before an action runs rather than after, an approach intended to limit unauthorized or destructive operations, and the system can resume work after an infrastructure failure. Organizations can also connect their own tools through the Model Context Protocol, so the assistant is not limited to functions Teradata ships itself.
The additions arrive as businesses try to give AI agents enough information and authority to complete work without losing control of data access and approvals. Sumeet Arora, Teradata's chief product officer, said 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. Teradata is also making a case on operating costs. In company-reported testing on the SWE-bench Pro benchmark using the same Opus 5 model, Tera used 73% fewer tokens than Claude Code, completed tasks 42% faster and had 58% lower total cost, while achieving a higher task completion rate. On the data-eng-bench pipeline engineering benchmark, Teradata reported a 53% lower cost per reliably solved task than Snowflake Inc.'s Cortex Code, based on Snowflake's published results. The company cautioned that the comparisons reflect its own testing and stated benchmark conditions, not performance across all enterprise workloads.
What this means for business
For companies adopting AI agents, the practical change is that governance moves from a policy document into the execution path. A request can be checked against access rules before an action runs, and a sensitive step can wait for a human approval instead of proceeding automatically. That matters most in regulated work — finance, healthcare, telecom — where an agent that reads customer data needs a traceable answer to where its output came from. A small company will likely notice the reusable Agent Skills first, because writing SQL or Python, optimizing queries, tuning workloads and sizing compute resources are tasks it may not have dedicated staff for. A large enterprise will care more about the Context Engine, since its problem is usually fragmented definitions and policies across many systems rather than a lack of tooling.
Several things remain to be verified before a decision. The benchmark figures are vendor-reported and explicitly tied to stated conditions, so they describe Tera against Claude Code and Cortex Code in those tests, not in a specific company's environment. The fourth-quarter 2026 availability date means the capabilities are not generally deployable today, and the source does not specify pricing or which models customers may choose beyond the company's statement that they can select models and run workloads in cloud, on-premises or sovereign environments. Teradata is offering AI Services to help customers identify use cases, configure Industry Knowledge Models and move enterprise knowledge into production faster, which suggests deployment still requires configuration work rather than a switch. Buyers should ask how lineage and access policies are enforced when an agent reaches data outside Teradata, what the human-approval step looks like in practice, and how token and cost measurements were produced.
The marker to watch is whether the fourth-quarter 2026 release ships with the Context Engine, Harness and Agent Skills as described, and whether customers publish their own cost and completion figures rather than repeating the vendor's benchmarks. If independent deployments confirm lower token use and cost per solved task under governed execution, the argument shifts from model quality to control over how agents act — and that is the ground on which data platform vendors will compete for enterprise AI budgets.
