The tech industry has been in a heated race over the past year to develop AI agents, with the spotlight frequently on their potential to transform workflows. However, researchers from Anthropic, Barry Zhang and Mahesh Murag, argue that the key to enhancing the effectiveness of AI in job functions may not lie in creating more sophisticated agents but rather in developing “agent skills.” This perspective was shared during their presentation at the AI Engineering Code Summit last month.
Zhang emphasized that the diversity of agents across different domains might not be as pronounced as previously believed. He suggested that instead of crafting unique agents for every scenario, organizations could benefit more from a singular, general-purpose agent that draws upon a comprehensive library of skills. “Skills are organized collections of files that package composable procedural knowledge for agents,” he explained. These skills essentially serve as folders containing everything an agent requires to perform tasks reliably and efficiently.
Despite the advancements in AI, Zhang pointed out that current agents often fall short in providing the necessary expertise and context for real-world applications. This gap is where skills become invaluable, equipping agents with domain-specific knowledge and standardized workflows. Murag noted that a remarkable array of skills has emerged since their launch, with contributions from professionals in fields such as accounting, legal, and recruiting. In just five weeks, thousands of skills have been created, prompting large corporations to begin viewing them as internal AI playbooks.
Major companies within the Fortune 100 are already utilizing these skills to “teach agents about their organizational best practices,” Murag revealed. The emergence of AI agents has been heralded by tech leaders as a groundbreaking transformation for workplace dynamics. OpenAI’s CEO Sam Altman highlighted that AI agents are already taking on tasks traditionally carried out by entry-level employees. At a recent summit, he articulated that workers are beginning to assign tasks to agents, evaluate their performance, and provide feedback, paralleling the management of a team of junior staff.
Altman also forecasted that as AI agents evolve, they might assist in uncovering fresh insights and tackling complex business challenges. Similarly, Microsoft’s AI platform product lead, Asha Sharma, predicted that AI agents could lead to a flattening of corporate structures, altering organizational dynamics significantly over the next few years by reducing the number of hierarchical layers.
However, some industry experts caution against the overhyping of AI agents. Guido Appenzeller, a partner at a16z, raised concerns during a recent podcast, suggesting that several startups may merely be enhancing traditional software with a chat interface, labeling it as an AI agent to justify higher pricing. “There’s a marketing angle to agents,” he stated, indicating that not all innovations in this space may genuinely revolutionize work processes.
As companies continue to explore the potential of AI agents and skills, the dialogue surrounding their role in the future of work remains pivotal. The discussions reflect a broader trend toward redefining workplace efficiencies and the tools necessary to achieve them.
