AI agents "will come of age in 2025, " according to AI futurist Steve Brown. Conversely, AI expert Tobias Zwingmann asserts that this year, "the rapid adoption of AI agents will not yield the productivity increases that many anticipate. " These perspectives, offered by two leading experts in the field, actually complement one another more than they clash. Business leaders are eager to determine if their companies can enhance operations while reducing costs. Economists, on the other hand, seek to know if these improvements will occur at a scale sufficient to be reflected in productivity statistics. AI agents embody my earlier prediction that most business functions cannot be enhanced merely by using chatbots connected to large language models. Rather, significant advancements are expected to emerge from specialized applications designed to address specific business challenges. Brown presents various examples of AI agents, some of which are already operational. A straightforward illustration is the Customer Support Agent: “An agent that handles customer support calls, schedules appointments, addresses inquiries, sells spare parts, and manages follow-ups, such as sending emails with product instructions. ” He also mentions customer support agents for travel and shopping, enterprise-based agents like purchasing assistants and scientific researchers, along with “ambient agents” capable of monitoring various facets of homes, public areas, or cybersecurity efforts. A crucial aspect of these agents is their specialization in particular domains. For instance, a customer service agent would be tailored to a specific company, equipped with knowledge of its products, technical specifications, and service call management systems. While a framework for these agents could be shared among different companies, each would require customization to align with individual operational needs. Zwingmann highlights that agents are far from plug-and-play solutions. “They perform impressively in demonstrations and controlled settings, but deployment often leads to unexpected complications, ” he notes.

He offers three key recommendations for implementing AI agents by 2025. The first is to start with small-scale projects, gradually increasing the agent's autonomy and decision-making power. Next, he advises prioritizing tasks, focusing on “specific, well-defined activities where the consequences of failure are manageable. Document processing is a great starting point, whereas high-stakes financial decisions might be better left for later. ” Lastly, Zwingmann emphasizes the importance of planning for failures. Agents will inevitably make errors, particularly in unusual scenarios or when circumstances change. Therefore, the implementation strategy should minimize risks and ensure that human intervention is readily available. I believe Brown would agree with these recommendations. As a futurist, he envisions the tremendous potential of AI agents, stating, “The near future of work will involve human, digital, and robotic employees collaborating closely to achieve objectives, each contributing unique strengths and weaknesses. Companies that manage to strike the right balance among their workforce, foster a culture of trust between human and digital employees, and leverage machines and robots to enhance the capabilities of their human staff will excel in the marketplace. ” Many successful companies have emerged from the partnership of two individuals with contrasting personalities: a bold visionary alongside a meticulous technician. One foresees a transformative future driven by radical change, while the other ensures the timely payment of bills and seeks to optimize production costs. The two articles addressing AI agents and the future of productivity mirror these distinct personalities. One explores a future where individuals are significantly empowered through the use of effective tools, while the other outlines a tangible, cost-effective pathway to realizing that vision.