Working alongside AMD, Intel, and NVIDIA as part of the MLCommons® taskforce, Workato helps define how the industry measures AI systems' ability to support agents in production

PALO ALTO, Calif.: Workato, the leading Control and Execution Platform for Enterprise AI, today announced its contribution to the MLPerf Agentic Inference Benchmark, the industry-standard AI benchmark suite from MLCommons. As a member of the MLPerf Inference Agentic taskforce, Workato – alongside taskforce representatives from AMD, Intel, and NVIDIA – contributed 500 real-world agentic workflow trajectories, grounded in its experience orchestrating enterprise business processes at scale.

The MLPerf Agentic Inference Benchmark is the first agentic AI infrastructure benchmark to measure performance against the complexity of real enterprise business processes. It is the first benchmark in MLPerf to measure multi-turn AI agents - how large language models (LLMs) are used in production. Rather than scoring a single question and answer, it measures how AI systems perform across the long, multi-step work real agents do: reading a request, calling tools, interpreting results, asking follow-up questions, and continuing until the task is resolved.

"As AI moves from the edge of the business to its core, the orchestration that directs and governs agent work becomes as critical as the models themselves," said Adam Seligman, Chief Technology Officer and General Manager of the AI Lab at Workato. “By grounding this benchmark in the kind of complex business processes Workato orchestrates in production every day, we're helping give model providers an independent standard for whether their AI agents can run reliably and at scale."

Workato’s contribution reflects one of the most common and demanding patterns in enterprise AI: real end-to-end business processes across databases, business systems, knowledge bases, and more. Its trajectories represent interactions in which a customer asks for help and an agent uses tools to retrieve orders, track shipments, check policies, escalate cases, and resolve account questions - the kind of multi-step, tool-driven work Workato orchestrates in production every day.

The trajectories are synthetic but modeled on Workato’s real-world experience running these agents for enterprise customers. Combined with agentic coding trajectories contributed by the broader taskforce, the dataset spans the full range of real agentic behavior.

This work reflects a deliberate division of expertise. AMD, Intel, and NVIDIA anchor the taskforce with AI systems and hardware leadership; Workato brings the enterprise orchestration domain knowledge that determines whether a benchmark reflects production reality. As the platform trusted by more than half of the Fortune 500 to orchestrate work across their core systems, Workato sees the multi-step, tool-driven, governed work that defines agentic AI in the enterprise - spanning finance, sales, IT, HR, and customer support, and acting across more than 14,000 connected applications.

That expertise is increasingly backed by formal research. Through its AI research labs in San Francisco and Singapore, Workato advances the science of autonomous enterprise agents, with work spanning synthetic evaluation, reinforcement learning, and customer-specific model optimization. Contributing to MLPerf extends that research into an open, reproducible, industry-wide standard.

A detailed technical breakdown of Workato’s contribution, including how the workflow trajectories stress AI serving systems, is available on the Workato blog.

The Agentic Inference benchmark will be available through the MLPerf Endpoints framework, with reference implementations, datasets, and accuracy thresholds published by MLCommons.

About Workato

Workato is the leading Control and Execution Platform for Enterprise AI - the neutral platform enterprises trust to put AI to work across their business. Workato unifies data, applications, and processes into a single platform so AI can reliably orchestrate business processes in production at enterprise scale. Built on more than a decade of running mission-critical processes for over half the Fortune 500 - including Nasdaq, Amazon, Cisco, Vodafone, Atlassian, and Lucid Motors - Workato turns over 14,000 enterprise systems AI needs to act on into one governed execution layer. For more information, visit workato.com.

About MLCommons and MLPerf

MLCommons® is an open engineering consortium that produces MLPerf®, a widely adopted suite of industry-standard benchmarks for measuring machine learning performance. MLPerf benchmarks are developed collaboratively by member organizations spanning industry and academia to provide fair, reproducible, and representative measurements of AI systems.

Source: Workato

Fonte: Business Wire