Teradata Enhances AI Workspace with Cost-Optimized Agent Execution
Teradata has announced significant upgrades to Tera, its AI-powered workspace for enterprise data and analytics tasks. The additions include a context engine, execution layer (Tera Harness), and reusable agent skills—all designed to make multistep workflows more efficient while controlling costs.
The new features address a growing concern among CIOs as they scale agentic workloads: the tendency for agents to burn tokens on unproductive loops. Tera’s enhancements aim to reduce unnecessary model calls by:
- Creating execution plans before sending tasks to LLMs
- Batching independent steps
- Dropping calls that don’t advance progress
Performance Gains and Cost Savings
According to Teradata’s internal evaluations using the SWE-bench Pro benchmark, Tera with these new capabilities demonstrated:
- 73% fewer tokens used compared to Claude Code
- 42% faster task completion
- 58% lower total cost while maintaining higher success rates
These improvements were achieved using the same Opus 5 model, highlighting the efficiency gains from optimized execution rather than simply choosing a cheaper model.
Implications for Enterprises
The focus on agent economics could become increasingly critical as enterprises expand their AI initiatives. By constraining how agents execute tasks, Teradata’s approach offers:
- Improved cost predictability: Consistent task completion times and resource usage make budgeting easier
- Reduced development effort: Platform-level optimizations mean less custom coding for each workflow
- Enhanced governance: Control over execution paths helps ensure responsible AI practices
While these benefits are compelling, CIOs should note that cost optimization comes with tradeoffs. For example:
- Pruning unnecessary steps could occasionally lead to incomplete or inaccurate results if critical information is inadvertently dropped
- Enterprises will need to define clear success criteria and review agent outputs to ensure quality
Overall, Teradata’s enhancements represent a practical approach to scaling agentic AI by focusing on execution efficiency rather than just model selection.