AMD released the TraceLens Agent on September 30, 2026, saying it automates performance bottleneck identification in AI training and inference workloads. It is the company's first software update since the release of the TraceLens library in 2025.
AMD reported the TraceLens Agent processes profiler traces from PyTorch, JAX, and AMD RocProf, analyzing each kernel's contribution to runtime and achieved FLOPs and bandwidth. That compares with the earlier release of the TraceLens library, which provided foundational performance analysis capabilities.
The TraceLens Agent is built on the TraceLens library and targets AI training and inference workloads. Availability begins with open-source release on GitHub, initially for developers and researchers.
"The TraceLens Agent turns a raw PyTorch profile into a prioritized, evidence-backed optimization report," said Steven K. Reinhardt, AMD's senior director of software tools. The agent streamlines the workflow for engineers to identify performance gaps and implement optimizations.
The announcement follows the release of the TraceLens library in 2025. AMD frames the significance of the TraceLens Agent as a step toward more efficient AI model development and deployment, adding no judgement of your own.
AMD did not say how widely the TraceLens Agent will be adopted, and it raised the open question of how effectively the agent will handle complex workloads. The source says the agent will support both standalone and comparative analysis on traces.
Source: [amd](https://rocm.blogs.amd. com/software-tools-optimization/tracelens-analysis-agent/README.html)
