In this episode of AI Explained, Josh Rubin from Fiddler AI hosts Nate Jones to discuss the current state of agentic AI adoption in enterprises, addressing the hype, hopes, and hidden risks. They define AI agents as large language models combined with tools and guidance, and explore the challenges organizations face when moving from initial excitement to successful implementation. Nate points out that many companies are in the "trough of disillusionment" due to the complexity of integrating agents into existing workflows, emphasizing the importance of problem framing, understanding data, and focusing on cultural change within the business. The conversation covers architectural decisions, the value of multi-agent systems versus simpler, more constrained approaches, and the need for robust instrumentation and ongoing evaluation to ensure AI agents perform reliably in production. They also touch on the build versus buy dilemma, with Nate favoring buying tools that empower developers but being more skeptical of fully готовых agent solutions due to the complex business contexts involved. The episode concludes with a discussion on digital twins, the role of unstructured data, and the challenges of data privacy in AI implementation, followed by audience questions.
Sign in to continue reading, translating and more.
Continue