On October 21, Beijing time, NVIDIA founder and CEO Jensen Huang said at the GTC Berlin conference that the development of AI will increase the number of users of software tools, and tool companies will also transform into AI companies and agent companies.
In his speech, Huang mentioned that some people believe the arrival of AI means the end of software and tool companies, and he holds the opposite view. He believes that as AI agents take on more work, the software tools developed in the past will see more users.
The reason supporting this judgment is that AI still needs to call on professional tools to complete actual work. Huang used the work of scientists and engineers as an example, noting that even with strong intelligence, they still need laboratories, simulators, and solver tools. AI agents likewise need tools to conduct experiments, simulations, and data processing.
He said that programming is one of the areas where agents landed early, and a compiler or a Python environment can support some of those tasks. As agents enter more industries, the tools they need to call will become richer, including fluid dynamics, quantum chemistry, and industrial design software.
A demo shown at the conference illustrated this way of working. In a demo related to Dassault Systèmes, an AI agent adjusted the door design of the Volkswagen Golf for use on the Tiguan according to engineering requirements, and called design and simulation tools to test the solution. After the simulation revealed a vibration problem, the design agent continued to make modifications, and engineers finally reviewed and selected the solution.
According to Huang's judgment, tool companies will in the future provide both software and agents that are specially optimized and skilled at using that software.
This judgment is also an important basis for NVIDIA's business expansion. In his speech, Huang repeatedly introduced its models, computing platforms, and software libraries, hoping to provide computing support for agents calling professional tools. The live demo presented this technical path, but did not provide cost and efficiency data after large-scale deployment.
Related reading: