AI agents move from assistants to owners of engineering workflows
AI agents in engineering design are software systems that autonomously run long, multi-step workflows across tools such as electronic design automation and molecular modeling, using physics-accurate engines to iteratively propose, test, and refine designs with minimal human supervision while still maintaining verification against deterministic simulators and rule-based flows.
The key shift is that AI agents are no longer sidecar copilots; they are becoming the primary operators of complex design flows. Siemens’ acquisition of Defacto Technologies and its expanded AI partnership with NVIDIA show this change most clearly, embedding agents into the heart of SoC and EDA AI automation workflows. At the same time, Schrödinger’s Bunsen points in the same direction for physics-based drug design AI, signaling a broader pattern: domain-specific, physics-aware agents are taking over the tedious yet critical iteration loops that have consumed engineers and scientists for decades.
SoC design automation: Siemens bets on RTL agents, not more engineers
Siemens’ move to acquire Defacto Technologies is an explicit bet that the future of SoC design automation lives at the RTL and assembly levels, where AI agents can make the biggest dent in iteration cycles. Defacto develops tools for SoC assembly, design restructuring and power intent management, all tied together with a unified, persistent design database that is tailor-made for automated workflows.
As chip makers shift from monolithic dies to chiplet-heavy, configurable SoCs, the explosion of IP blocks and interconnects has made manual SoC assembly slow and error-prone. According to Siemens, manual RTL processes can no longer keep up, and the answer is early-phase automation that spans connectivity, timing constraints and power intent flows in one continuous pipeline. The opinionated takeaway: Siemens does not expect to scale design teams linearly with complexity. Instead, it plans to scale AI agents that can reason over RTL structure, restructure designs on the fly, and feed cleaner, verified intent downstream.
EDA AI automation: NVIDIA-powered agents as always-on verification teams
If Defacto is about feeding agents cleaner SoC structures, Siemens’ expanded partnership with NVIDIA is about giving those agents more brains and faster physics. Siemens has extended its Fuse EDA AI Agent system with NVIDIA Nemotron models, CUDA-X libraries, NeMo Gym, and OpenShell to support self-verifying agentic AI workflows for semiconductor and PCB design.
These AI agents can now reason, act and validate decisions against deterministic, physics-based EDA engines, and orchestrate end-to-end workflows across synthesis, verification, implementation and validation. In practice, this means NVIDIA EDA workflows where long-running agents tweak designs, call EDA tools, read back results and refine strategies in a loop, all while maintaining audit trails and access control via OpenShell. The most telling detail is that library characterization workflows using Siemens’ Solido suite now automate Liberty file generation and verification, cutting characterization turnaround by more than 10x and reducing token costs by 5x to 10x. That is not a productivity nudge; it is a structural change in how verification is done.
From SoCs to molecules: agents kill the iteration tax
The pattern across hardware and molecular design is clear: AI agents are being wired directly into physics engines so they can automate the most painful iteration loops. In SoC flows, Siemens and Defacto aim to eliminate manual SoC assembly and restructuring by letting agents manage connectivity, timing and power intent over a persistent database. In EDA verification, agents now handle repetitive tasks like library characterization and layout analysis, including natural-language-driven result analysis, fix suggestions and report generation.
Schrödinger’s Bunsen fits the same mold for physics-based drug design AI, where agents will operate over proprietary molecular modeling software to run design–simulate–evaluate cycles. The bottom line is that engineers and scientists will increasingly supervise AI agents instead of manually pushing designs through tools. The organizations that win will be those that treat agent design as a first-class engineering discipline, not a bolt-on feature to existing tools.
What changes for engineering teams next
The near-term future is not about a single omniscient AI, but fleets of domain-specific agents wired into deterministic engines. Siemens’ Fuse EDA AI Agent already coordinates multi-agent workflows, with each agent scoped to tasks like synthesis, verification or custom IC layout for semiconductor and PCB design. Intelligence Center X then orchestrates these agents across design, manufacturing and supply chain, extending digital twin environments.
For ordinary users, this means fewer late surprises and shorter time-to-results. Long-running verification runs become continuous conversations with AI agents that spot issues earlier and help close verification faster. The expanded AI-driven EDA features are slated to arrive in upcoming releases of Siemens’ AI-native portfolio, while Defacto’s integration promises more consistent SoC design lifecycles end to end. The conclusion is blunt: if your engineering process still relies on manual iteration across disconnected tools, you are competing against teams whose AI agents are already working overnight—inside the same physics engines you use today.
