From OpenAI to Nvidia, researchers agree: AI agents have a long way to go

Only in the Bay Area does spending a Saturday geeking out about AI agents—alongside 2,000 students, researchers, and tech insiders crammed into UC Berkeley—feel like a totally normal weekend plan. At the day-long Agentic AI Summit, the atmosphere was less like an academic conference and more like Silicon Valley’s buzzy hotspot. The speaker lineup was impressive, featuring Jakob Pachocki, chief scientist at OpenAI; Ed Chi, VP of research at Google DeepMind; Bill Dally, chief scientist at Nvidia; Ion Stoica, cofounder at Databricks & Anyscale and UC Berkeley professor; and Dawn Song, a UC Berkeley professor specializing in AI security. The summit’s focus was on AI agents—AI-powered systems that can autonomously complete tasks using other software tools. Imagine a chatbot that not only suggests a vacation itinerary but also books flights and hotel reservations. As my colleague Jeremy Kahn noted in a recent article, this kind of automation has long been a C-suite dream. Over the past decade, companies have used robotic process automation (RPA) to automate repetitive tasks like data entry. However, traditional RPA is inflexible and limited to narrow tasks. Agentic AI aims to be more flexible and powerful, adapting to complex business needs. OpenAI CEO Sam Altman expressed optimism in a January 2025 blog post, saying, “We believe that, in 2025, we may see the first AI agents ‘join the workforce’ and materially change the output of companies.” However, despite the hype, the overall message at the summit was cautious. AI agents remain unreliable and often fail to remember previous interactions. Ed Chi from Google DeepMind emphasized the gap between impressive demos and real-world production readiness. Jakob Pachcki highlighted concerns about safety, security, and trustworthiness, especially when agents operate autonomously in sensitive areas. Sherwin Wu, head of engineering at OpenAI API, remarked, “I still don’t think agents have really lived up to their promise. Certain more generic cases have worked, but my day-to-day work doesn’t really feel that different with agents.” Though today’s AI agents may not yet fulfill the massive expectations—contrasting with Salesforce CEO Marc Benioff’s recent claim that digital labor will transform leadership—the summit speakers remained optimistic. Ion Stoica pointed to infrastructure improvements easing agent development. Bill Dally from Nvidia noted that hardware advances will enable more powerful and efficient agents. Several experts cited “narrow wins” in specific domains such as coding. Despite current limitations, the packed UC Berkeley ballroom reflected the industry’s focus on achieving AI agents capable of reliable real-world operation. The potential payoff, they believe, will be worth the wait.

---

AI in the News

• U.S. agency approves OpenAI, Google, Anthropic for federal AI vendor list. The General Services Administration added OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude to an approved AI vendor list to accelerate government adoption. These providers commit to responsible use and federal compliance.
• AI spending boom impacts U.S. economy. Big Tech’s $350 billion AI investment in 2025 is fueling data center construction and chip demand, potentially boosting GDP growth by up to 0.7%. Economists warn of risks if the AI boom slows.
• Sales tool Clay raises $100 million at $3.1 billion valuation. Clay, which helps sales reps find leads, raised funds led by CapitalG, Alphabet’s investment arm.

Eye on AI Research

• Google DeepMind’s Genie 3 'world model' creates real-time interactive simulations. Genie 3 generates dynamic virtual worlds from text prompts, enabling navigation at 24 frames per second. It represents a step toward AI systems that understand and simulate real-world environments, crucial for advanced agents and general AI.

---

This article was originally published by Sharon Goldman at Fortune on August 5, 2025.

FAQ

What are AI Agents?

AI agents are AI-powered systems designed to autonomously complete tasks by interacting with other software tools. Think of them as sophisticated digital assistants that can go beyond simple requests to execute complex actions.

What are the current limitations of AI Agents?

Despite their potential, current AI agents often face limitations such as unreliability, an inability to consistently remember past interactions, and a gap between impressive demonstrations and real-world production readiness. Concerns also exist around safety, security, and trustworthiness, particularly when agents operate autonomously in sensitive domains.

What is the future outlook for AI Agents?

Industry experts remain optimistic about the future of AI agents. They anticipate that ongoing infrastructure and hardware advancements will lead to more powerful and efficient agents. While current capabilities are focused on "narrow wins" in specific areas like coding, the industry is striving towards agents that can operate reliably in real-world scenarios.

What is the difference between traditional RPA and Agentic AI?

Traditional Robotic Process Automation (RPA) is designed to automate repetitive, narrowly defined tasks. Agentic AI, on the other hand, aims to be more flexible and powerful, capable of adapting to complex business needs and handling a broader range of tasks autonomously.

When can we expect AI Agents to significantly impact the workforce?

Some industry leaders, like OpenAI CEO Sam Altman, have predicted that by 2025, AI agents might begin to "join the workforce" and measurably impact company output. However, the general sentiment from experts suggests that widespread, reliable integration still requires further development.

Crypto Market AI's Take

The insights from the Agentic AI Summit highlight a crucial phase in AI development. While the potential for AI agents to revolutionize task completion and automation is immense, the current reality underscores the need for continued research and development. At Crypto Market AI, we understand this duality. Our focus on AI-driven crypto trading bots and AI analysts aims to harness the power of AI for market intelligence while acknowledging the ongoing challenges. We strive to build robust systems that not only leverage AI's capabilities but also prioritize safety, reliability, and user trust, mirroring the industry's progression towards more dependable and sophisticated AI agents.

More to Read:

• AI-driven Crypto Trading Tools Reshape Market Strategies in 2025
• The Future of Cryptocurrency: What's Changing and Why It Matters
• Understanding AI Agent Washing: Risks and Realities