AMD and Fiddler have announced a collaboration aimed at improving the reliability, governance, and scalability of enterprise AI agents. By combining AMD's high-performance computing capabilities with Fiddler's observability and policy tools, the partnership seeks to address the growing complexity of deploying autonomous AI systems in production environments.

AMD’s High-Performance Compute Foundation

AMD provides the computational backbone for AI workloads through its Instinct™ GPUs and ROCm™ open software platform. These tools enable enterprises to train, fine-tune, and deploy AI models at scale. AMD’s open platform supports a range of deployment scenarios, allowing organizations to evaluate infrastructure performance and scalability against their specific requirements. This foundation is designed to meet the resource demands of modern AI applications, including agentic workflows that involve multiple model calls, document retrievals, and tool interactions.

Fiddler’s Observability and Governance Tools

Fiddler complements AMD’s compute capabilities with its AI Control Plane, which provides end-to-end observability for agentic systems. This includes monitoring application-level results, tracing individual spans, and evaluating risks such as hallucinations, harmful content, and sensitive data exposure. Fiddler’s tools allow teams to identify and address issues at specific steps in an agent’s workflow, such as incorrect document retrieval or inefficient tool usage. The platform also enforces governance policies through its Guardrails feature, screening prompts and responses for compliance with enterprise standards.

Addressing Challenges in Autonomous AI

As enterprises increasingly deploy AI agents that perform complex tasks, such as interpreting internal policies or automating workflows, the need for robust monitoring and governance becomes critical. Traditional monitoring systems, designed for simpler AI models, often fail to capture the nuanced failures of agentic workflows. For example, an agent might provide a response that conflicts with source policies, an issue that infrastructure metrics alone cannot detect. Fiddler’s observability tools bridge this gap by connecting application outcomes to the underlying model calls and tool interactions that produced them.

Practical Benefits for Enterprises

The collaboration between AMD and Fiddler offers several practical advantages for enterprise AI deployments:

  • Improved User Experiences: Continuous quality monitoring helps reduce errors and the need for manual corrections.
  • Faster Troubleshooting: Detailed traces and diagnostics enable teams to pinpoint and resolve issues more efficiently.
  • Cost Optimization: Visibility into token usage and agent activity helps identify and address expensive or inefficient workflows.
  • Enhanced Accountability: Fiddler’s unified agent registry and audit trails support compliance and internal reviews.
  • Flexibility: AMD’s open software foundation and Fiddler’s cloud-agnostic approach allow enterprises to adapt to evolving infrastructure and application requirements.

Building Reliable AI Systems

By integrating AMD’s compute platform with Fiddler’s evaluation and governance tools, enterprises can achieve a continuous feedback loop that ensures AI systems deliver dependable results at predictable costs. This partnership underscores the importance of combining high-performance infrastructure with robust oversight to scale autonomous AI responsibly.