Teradata aims to make agentic execution of multistep data work more efficient
InfoWorld ·
Teradata is adding a context engine, an execution layer, and reusable agent skills to Tera, its AI-powered workspace for enterprise data and AI tasks, in order to make agentic execution of multistep workflows more efficient. Tera was initially introduced in May as part of Teradata’s Autonomous Knowledge Platform. The new additions are designed to cut unnecessary model and tool calls while preserving business context and automatically matching each task with the right data, tools, models, and skills, helping enterprises control inference costs as agentic workloads scale, Teradata said in a statement . The new execution layer, Tera Harness, determines how agents approach tasks and how workflows are routed, while the Tera Context Engine adds the business context needed to guide those decisions. In order to reduce the computation needed to complete a task or workflow, the Harness creates an execution plan before sending work to an LLM, batches independent tasks, and drops model or tool calls that do not advance the task, the company said. It applies 84 execution patterns before inference and limits how many steps a workflow can run based on its progress, reducing repeated LLM reasoning and the token and infrastructure costs associated with unproductive agent loops, it added. According to Teradata’s own evaluations on the SWE-bench Pro benchmark, with these new capabilities Tera used 73% fewer tokens than Claude Code , completed tasks 42% faster, and incurred 58% lower total cost while achieving higher task completion rates, while running the same Opus 5 model . Managing the cost of agentic workloads That focus on the economics of agent execution could become increasingly important for CIOs trying to scale agentic workloads within their allocated AI budgets, analysts said. “Tera Harness is attacking the part of agentic AI that enterprises are only now discovering hurts, which is that an agent left to reason its way through every step will happily burn tokens on loops that never move the task forward. Even a small reduction in calls per workflow can become meaningful at scale,” said Ashish Chaturvedi , executive research leader at HFS Research. For enterprises looking to control agent costs, cutting unnecessary model calls and the reasoning tokens they consume could be more practical than simply choosing a cheaper model, said Stephanie Walter , practice lead of AI stack at HyperFrame Research, as it reduces the amount of inference an agent needs rather than simply reducing the cost of each inference. Beyond reducing spend, the same ability to constrain how an agent executes a task can also help CIOs budget for their agentic workloads better, according to Advait Patel , senior site reliability engineer at Broadcom. “Right now, agentic costs are hard to forecast because the same task can take five calls one day and fifty the next. A harness that plans first makes costs more consistent, which makes budgeting and scaling decisions easier,” Patel said. That predictability could also reduce the amount of agent-optimization work developers have to build and maintain themselves. “Planning before invoking an LLM, batching independent steps, and dropping calls that do not advance a task are established engineering practices, but having the platform apply them by default means individual development teams do not have to implement and continuously tune those controls for every agent workflow, Patel pointed out. Tera’s cost controls come with tradeoffs However, Tera’s cost-optimized approach is not without tradeoffs, and enterprise teams will need to account for the constraints that can come with more tightly controlled agent execution. “The catch is that pruning inference calls to an LLM is a judgment call. If the Harness drops a step it considers unnecessary and that step turns out to matter, the enterprise saves money but gets a worse answer. Enterprises will therefore need to verify the results carefully,” Patel said. Walter said that, in turn, could shift some of the work for developers from directing individual steps such as deciding which tools to invoke and how tasks should be sequenced to defining what a correct outcome looks like and reviewing whether an agent has achieved it. Teams will also need to maintain the skills, instructions, and guardrails that determine how agents operate in production, creating a new layer of engineering work even as the platform takes on more of the execution itself, she added. CIOs too will need to exercise caution, particularly around the promised cost efficiencies, said Robert Kramer , managing partner at KramerERP. “Teradata’s benchmark results should not be treated as equivalent to enterprise total cost of ownership,” he said. “Enterprises should instead measure the cost of completing a business task, including model usage, data compute, tool calls, retries, orchestration, and the human effort required to review the result.” And the cost of running agentic workflows is not the only consideration for CIOs, said Walter: The more enterprises rely on Tera’s architecture to manage context, execution, and reusable skills, the more those capabilities could become embedded in their workflows, raising a separate question around how easily those workflows could be moved to another platform. Early adopters Those tradeoffs and Teradata’s current install base are also likely to shape which enterprises choose to adopt Tera, and how. “The most likely early adopters are existing Teradata customers with complex, governed data environments and workflows spanning analytics, data engineering, and AI. For those enterprises, Tera represents a logical expansion of an environment they already use,” Walter said. “Winning customers that have standardized on Snowflake or Databricks will be harder because the agents are not a good enough reason for lift and shift.” Teradata plans to make the new context engine, execution layer, and reusable agent skills generally available by December. This article first appeared on CIO .
Teradata is adding a context engine, an execution layer, and reusable agent skills to Tera, its AI-powered workspace for enterprise data and AI tasks, in order to make agentic execution of multistep workflows more efficient. Tera was initially introduced in May as part of Teradata’s Autonomous Knowledge Platform. The new additions are designed to cut unnecessary model and tool calls while preserving business context and automatically matching each task with the right data, tools, models, and skills, helping enterprises control inference costs as agentic workloads scale, Teradata said in a statement . The new execution layer, Tera Harness, determines how agents approach tasks and how workflows are routed, while the Tera Context Engine adds the business context needed to guide those decisions. In order to reduce the computation needed to complete a task or workflow, the Harness creates an execution plan before sending work to an LLM, batches independent tasks, and drops model or tool calls that do not advance the task, the company said. It applies 84 execution patterns before inference and limits how many steps a workflow can run based on its progress, reducing repeated LLM reasoning and the token and infrastructure costs associated with unproductive agent loops, it added. According to Teradata’s own evaluations on the SWE-bench Pro benchmark, with these new capabilities Tera used 73% fewer tokens than Claude Code , completed tasks 42% faster, and incurred 58% lower total cost while achieving higher task completion rates, while running the same Opus 5 model . Managing the cost of agentic workloads That focus on the economics of agent execution could become increasingly important for CIOs trying to scale agentic workloads within their allocated AI budgets, analysts said. “Tera Harness is attacking the part of agentic AI that enterprises are only now discovering hurts, which is that an agent left to reason its way through every step will happily burn tokens on loops that never move the task forward. Even a small reduction in calls per workflow can become meaningful at scale,” said Ashish Chaturvedi , executive research leader at HFS Research. For enterprises looking to control agent costs, cutting unnecessary model calls and the reasoning tokens they consume could be more practical than simply choosing a cheaper model, said Stephanie Walter , practice lead of AI stack at HyperFrame Research, as it reduces the amount of inference an agent needs rather than simply reducing the cost of each inference. Beyond reducing spend, the same ability to constrain how an agent executes a task can also help CIOs budget for their agentic workloads better, according to Advait Patel , senior site reliability engineer at Broadcom. “Right now, agentic costs are hard to forecast because the same task can take five calls one day and fifty the next. A harness that plans first makes costs more consistent, which makes budgeting and scaling decisions easier,” Patel said. That predictability could also reduce the amount of agent-optimization work developers have to build and maintain themselves. “Planning before invoking an LLM, batching independent steps, and dropping calls that do not advance a task are established engineering practices, but having the platform apply them by default means individual development teams do not have to implement and continuously tune those controls for every agent workflow, Patel pointed out. Tera’s cost controls come with tradeoffs However, Tera’s cost-optimized approach is not without tradeoffs, and enterprise teams will need to account for the constraints that can come with more tightly controlled agent execution. “The catch is that pruning inference calls to an LLM is a judgment call. If the Harness drops a step it considers unnecessary and that step turns out to matter, the enterprise saves money but gets a worse answer. Enterprises will therefore need to verify the results carefully,” Patel said. Walter said that, in turn, could shift some of the work for developers from directing individual steps such as deciding which tools to invoke and how tasks should be sequenced to defining what a correct outcome looks like and reviewing whether an agent has achieved it. Teams will also need to maintain the skills, instructions, and guardrails that determine how agents operate in production, creating a new layer of engineering work even as the platform takes on more of the execution itself, she added. CIOs too will need to exercise caution, particularly around the promised cost efficiencies, said Robert Kramer , managing partner at KramerERP. “Teradata’s benchmark results should not be treated as equivalent to enterprise total cost of ownership,” he said. “Enterprises should instead measure the cost of completing a business task, including model usage, data compute, tool calls, retries, orchestration, and the human effort required to review the result.” And the cost of running agentic workflows is not the only consideration for CIOs, said Walter: The more enterprises rely on Tera’s architecture to manage context, execution, and reusable skills, the more those capabilities could become embedded in their workflows, raising a separate question around how easily those workflows could be moved to another platform. Early adopters Those tradeoffs and Teradata’s current install base are also likely to shape which enterprises choose to adopt Tera, and how. “The most likely early adopters are existing Teradata customers with complex, governed data environments and workflows spanning analytics, data engineering, and AI. For those enterprises, Tera represents a logical expansion of an environment they already use,” Walter said. “Winning customers that have standardized on Snowflake or Databricks will be harder because the agents are not a good enough reason for lift and shift.” Teradata plans to make the new context engine, execution layer, and reusable agent skills generally available by December. This article first appeared on CIO .