Industry Observations on AI Agents: Opportunities and Challenges

The year 2026 is being hailed as the year of artificial intelligence agents, with AI agents becoming remarkably ubiquitous in enterprise environments. They are utilized for a variety of tasks, including writing code, communicating with customers, and executing complex operations. However, the inherent risks associated with agentic behavior are also making headlines, as stories of agents malfunctioning and causing issues are becoming more common. According to SiliconANGLE, the greatest strength and weakness of AI agents lies in their nondeterministic behavior. Today's AI agents typically rely on large language models (LLMs), which can produce varying outputs for the same input due to their training methods. This nondeterminism grants agents their unique capabilities, allowing them to make decisions based on available information. However, it also means that in scenarios where precise outcomes are critical, such as payroll processing or healthcare, the use of AI agents could lead to disastrous consequences. Moreover, the costs associated with deploying AI agents can be significant, and organizations must carefully evaluate whether the benefits outweigh these costs. The New Stack further emphasizes that the effectiveness of AI agents is contingent upon the underlying infrastructure; without proper context, API calls can stall, leading to diminished performance. Therefore, from an industry perspective, businesses should approach the deployment of AI agents with caution, considering both their potential and the risks involved.