Agentrys bets on workflow-level chip design AI

Mark Ren, a former IBM and Nvidia design automation researcher who led Nvidia’s ChipNeMo project, argues that access to frontier AI alone will not create lasting semiconductor advantage. Agentrys, the startup he now leads, is built around the view that common models can raise productivity across the industry while leaving competitors with similar capabilities.

Agentrys focuses on automating chip-design workflows rather than isolated tasks. Ren says verification, physical implementation, timing closure, analog layout, and signoff require systems that understand customer-specific methods, preserve traceability, expose design state, and provide structured feedback for autonomous agents. The company calls its approach Agentic Design Automation, combining agents, engineering knowledge, specialized tools, evaluation, and continuous learning inside a customer’s own environment.

The platform is designed to keep sensitive design data, workflow history, acceptance criteria, configurations, and learned adaptations under customer control, including private-cloud and air-gapped setups. Agentrys says its Self-Evolve framework uses each workflow execution to refine agents and processes against grounded references and customer-defined criteria.

Agentrys has demonstrated an autonomous multi-agent flow that took a small 32-bit CPU from specification to sign-off-clean GDS layout with no human in the loop, using open-source tools and closing with zero routing violations. The company is engaging customers in verification and physical design, with RTL, analog, and system design on its roadmap.