Overview
- At GTC Berlin Jensen Huang rejected the idea that AI will end software and tool companies, saying agents will rely on professional tools rather than replace them.
- He identified programming as an early domain for agents, noting that compilers and Python environments can let agents carry out concrete coding tasks.
- A Dassault Systèmes demonstration showed an agent modify a Volkswagen Golf door design for a different model, run simulations that found vibration problems, iterate the design and leave final selection to engineers.
- NVIDIA framed its models, GPU compute and software libraries as the underlying infrastructure to let agents call specialized simulation, design and scientific tools.
- The presentations did not provide large-scale cost, efficiency or operational metrics, leaving open questions about deployment timelines, ROI and how companies will scale agent-driven workflows for routine engineering work.