In the bustling Bay Area, UC Berkeley recently played host to over 2,000 tech enthusiasts at the Agentic AI Summit—a meeting that was as much about innovation as it was about meeting like-minded people. With voices from OpenAI, Google DeepMind, and Nvidia filling the room, the focus was squarely on AI agents, systems designed to work independently with a variety of software tools.
The demos were impressive, but many experts agreed that these agents still face significant hurdles when it comes to everyday applications. Jakob Pachocki of OpenAI and Ed Chi from Google DeepMind noted that what shines in controlled settings often struggles in the real world. Sherwin Wu of OpenAI API summed it up candidly: “I still don’t think agents have really lived up to their promise.” This honest admission resonated with many at the event.
Not all was bleak, though. Improvements in infrastructure and hardware were highlighted by Ion Stoica from Databricks and Bill Dally from Nvidia. They suggested that these advances could soon make AI systems more robust and efficient. Meanwhile, major players like Google and Microsoft continue to drive large-scale investments in AI—a trend that has its perks but also carries risks, especially if the rapid pace of progress ever slows down.
The landscape is shifting fast. The U.S. General Services Administration has just approved AI tools from OpenAI, Google, and Anthropic for federal use, indicating growing trust in these emerging technologies. Although AI agents haven’t yet reached their full potential, the cautious optimism of the experts hints that as software and hardware advances merge, we might soon see these systems become a seamless part of our daily tech interactions.
