AI agents stop assisting and start engineering
NVIDIA’s AI agents for engineering are software systems that combine domain tools, physics models and accelerated computing libraries to automatically run simulations, interpret results and modify chip, system and quantum designs with minimal human intervention across complex workflows. NVIDIA is no longer pitching AI as a chatty helper; it is wiring agents directly into physics solvers, EDA stacks and quantum labs. The expansion of the NVIDIA Agent Toolkit adds PhysicsNeMo and CUDA-X libraries as agent-ready skills, turning AI into a programmable engineer for product design and development workflows. Siemens, meanwhile, is baking these agents into its Fuse EDA AI Agent to create self-verifying flows for semiconductor and PCB design. The message is clear: tedious engineering grunt work—from simulation sweeps to quantum computer calibration—is being handed off to AI agents that are expected to act, reason and check their own work.
PhysicsNeMo and CUDA-X: making simulation an AI-native skill
The most important change in the NVIDIA Agent Toolkit is the re-architecture of PhysicsNeMo into agent-friendly libraries and its pairing with new and updated CUDA-X components. PhysicsNeMo now provides AI physics tools that agents can call directly to train and deploy customizable models for design and simulation tasks, effectively turning numerical solvers and learned surrogates into callable functions inside an AI workflow. CUDA-X libraries add accelerated solvers and quantum chemistry capabilities, including iterative sparse solvers like cuISS and direct sparse solvers like cuDSS, which are vital for physics-based, electronic design and scientific simulations. The practical impact is less manual intervention in simulation-based engineering: AI agents can run parameter sweeps, refine models, and feed results back into chip design, verification, packaging and systems engineering, giving enterprises tighter control over customization, deployment and data privacy when building NVIDIA AI agents for engineering.
Ising Calibration: AI that tunes the quantum hardware itself
Quantum computer calibration has been a delicate art, but NVIDIA Ising Calibration turns it into an automated, agent-ready workflow. The open source vision language model interprets diagnostic outputs from quantum processors and decides how they should be tuned to keep operating, directly tackling the headache of quantum computer calibration. The latest release, Ising Calibration 1.5, can analyze unfamiliar diagnostic results without prior training examples and is 11.4% smaller at BF16 precision, making lab deployment more practical. It also uses examples from related experiments and is now 86.68% better than its predecessor when using such in-context examples. "Ising Calibration 1.5 advances AI and quantum computing calibration by outperforming all open models out of the box and on the QCalEval benchmark." With a 31‑billion‑parameter VLM that also ships in NVFP4-quantized form, labs can run agentic calibration workflows on a single GPU or NVIDIA DGX Spark, pushing quantum calibration toward fully automated bring-up and retune operations.
Siemens and NVIDIA: EDA AI workflows become self-verifying
While NVIDIA builds the toolkit, Siemens is putting it to work in production EDA AI workflows. The expanded strategic partnership adds Nemotron, OpenShell and CUDA-X support to the Fuse EDA AI Agent for verification, custom IC and PCB design. These agents combine Siemens’ engineering and verification engines with NVIDIA AI infrastructure, including NeMo Gym for optimizing long-running EDA agents, OpenShell for secure agent runtime, Nemotron and Switchyard for reasoning, and accelerated computing to cut simulation runtimes without sacrificing accuracy. Siemens is now integrating Fuse EDA AI Agent into Intelligence Center X to orchestrate agents across design, manufacturing and supply chains, extending the digital twin landscape. The payoff is tangible: automated library characterization workflows already reduce characterization turnaround times by more than 10x and lower token costs by 5x to 10x, while supporting advanced-node standard cell, memory and custom IP libraries. The expanded AI-driven EDA capabilities will arrive in upcoming releases of Siemens’ AI-native EDA portfolio.
From RTL coding to quantum chemistry: what engineers gain—and lose
Put together, these moves show NVIDIA AI agents engineering toward owning entire workflows rather than isolated tasks. Nemotron 3 Ultra now supports agentic RTL coding and ranks among open models for automated RTL coding on Verilog design problems, especially when paired with ACE-RTL, NVIDIA’s hardware design agent. PhysicsNeMo and CUDA-X extend that reach into quantum chemistry and physics simulation, while Siemens’ Fuse EDA AI Agent brings self-verifying flows into custom IC and PCB design. On the quantum side, the Ising Calibration blueprint ties Ising 1.5 to the NVIDIA NeMo Agent Toolkit, letting labs script full quantum calibration experiment automation. Engineers gain time and consistency: repetitive tasks like writing RTL, running characterization, or interpreting calibration plots are increasingly handled by agents. The trade-off is cultural, not technical—engineering teams must accept that the “doer” of many tasks is now software, and their role shifts toward defining constraints, validating edge cases and deciding when the agent’s output is good enough to trust.
