Teradata adds context engine and execution controls to Tera AI assistant

Teradata is adding three core capabilities to its Tera AI assistant that aim to let AI agents operate across enterprise data systems while staying within organizational access controls and approval workflows.

The additions, set for release in the fourth quarter of 2026, include a context engine that connects information from multiple sources, an execution system that manages multi-step tasks and pauses for human approval when needed, and a library of reusable data engineering functions. The updates were first reported by SiliconANGLE.

Tera is a natural-language interface that lets business analysts, data engineers, and database administrators interact with enterprise data without writing SQL or code. Users can request analysis, pipeline creation, or infrastructure management through conversational prompts. While Tera is part of Teradata's Autonomous Knowledge Platform, the company said it can access data held outside Teradata environments.

How the new components work

The Tera Context Engine connects metadata, data lineage, business definitions, and access policies from databases, catalogs, and pipelines without requiring data movement. The engine interprets requests based on organizational context and traces AI outputs back to their sources. Teradata said it applies consistent policies as information moves between systems.

Tera Harness handles execution by selecting appropriate tools, models, and data for each task. It tracks progress across multiple steps, can pause for human approval before sensitive actions, and resumes work after infrastructure failures. The system applies controls before actions run to prevent unauthorized or destructive operations.

Agent Skills packages common data tasks into reusable functions for writing SQL or Python, optimizing queries, tuning workloads, and sizing compute resources. Organizations can integrate their own tools through the Model Context Protocol.

Performance and cost claims

Teradata reported internal benchmark results comparing Tera to other AI coding systems. On the SWE-bench Pro benchmark using the Opus 5 model, the company said Tera used 73% fewer tokens than Claude Code, completed tasks 42% faster, and had 58% lower total cost while achieving higher task completion rates. On the data-eng-bench pipeline engineering benchmark, Teradata reported 53% lower cost per reliably solved task than Snowflake's Cortex Code, based on Snowflake's published results. The company noted these comparisons reflect its testing conditions and may not represent performance across all enterprise workloads.

"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," said Sumeet Arora, Teradata's chief product officer.

Why it matters

The challenge of giving AI agents enough autonomy to complete useful work without losing control over data access and approvals is becoming central to enterprise AI deployment. Teradata's approach embeds governance and approval checkpoints into the agent execution layer rather than treating them as external controls. This architecture matters because it attempts to solve the trust problem that prevents many organizations from moving AI agents beyond pilot projects into production workflows that touch sensitive data or make consequential decisions.

Teradata said customers will be able to choose their own models and run workloads in cloud, on-premises, or sovereign environments. The company is offering AI Services to help customers identify use cases and configure the business knowledge Tera needs for deployment.

Details were first reported by SiliconANGLE.