Anthropic researchers say the industry should stop building tons of AI agents — the real breakthrough is something simpler

In a rapidly evolving tech landscape, the focus on developing numerous AI agents has taken center stage over the past year. However, researchers from Anthropic suggest a more effective approach that centers on enhancing 'agent skills' rather than simply increasing the number of agents available. During a presentation at the AI Engineering Code Summit last month, Barry Zhang and Mahesh Murag emphasized that the real innovation lies not in creating diverse agents for every task but in cultivating a universal agent that can leverage a comprehensive library of skills. Zhang explained, 'We used to think agents in different domains will look very different. The agent underneath is actually more universal than we thought.' These skills represent organized collections of resources that equip agents with the procedural knowledge necessary to perform their tasks efficiently. Zhang pointed out that today's AI agents, despite their advanced capabilities, often lack the context and expertise required to excel in real-world applications. By integrating these skills, agents can gain essential domain knowledge and reusable workflows that enhance their effectiveness. Anthropic has already begun to witness the development of skills from individuals in various fields, including accounting, legal, and recruitment roles. In just five weeks since the skills' introduction, users have created thousands of them, with major corporations starting to utilize them as internal guidelines for their AI. Murag highlighted that Fortune 100 companies are employing these skills to educate agents about their best organizational practices. Tech leaders are optimistic about the potential of AI agents, viewing them as transformative tools for workplace efficiency. OpenAI CEO Sam Altman remarked that AI agents are already taking on responsibilities traditionally held by junior employees, indicating a shift in workplace dynamics. 'You hear people that talk about their job now is to assign work to a bunch of agents, look at the quality, figure out how it fits together, give feedback, and it sounds a lot like how they work with a team of still relatively junior employees,' he noted. Furthermore, Microsoft's Asha Sharma discussed the potential for AI agents to alter corporate structures, suggesting that the future may see a reduction in organizational layers. Nonetheless, the excitement surrounding AI agents does have its skeptics. Guido Appenzeller, a partner at a16z, cautioned against the overhype, noting that some startups might simply be rebranding basic language models as agents to justify higher pricing. 'There's a marketing angle to agents,' he stated, indicating a need for critical scrutiny in the industry. As the conversation around AI agents evolves, the focus on enhancing agent skills may just be the key to unlocking their full potential.