The tech industry has spent the past year racing to build AI agents, but Anthropic researchers say a simpler idea can make AI more effective on the job.
Barry Zhang and Mahesh Murag said at the AI Engineering Code Summit last month that the real breakthrough for agent workflow isn’t more agents, but “agent skills.”
“We used to think agents in different domains will look very different,” Zhang said in a clip of the talk published Monday. “The agent underneath is actually more universal than we thought.”
Instead of building new agents for every use case, companies should rely on a single general agent powered by a library of skills, Zhang said.
Skills are “organized collections of files that package composable procedural knowledge for agents,” Zhang said. They are simply folders that contain whatever an agent needs to complete a task consistently and efficiently.
Despite their intelligence, Zhang said today’s agents “lack expertise” and often miss important context in real-world use cases. Skills help fill in those gaps by giving agents domain knowledge and reusable workflows.
Murag said Anthropic has already seen skills built by people in accounting, legal, recruiting, and other non-technical roles. In the five weeks since launch, users have created thousands of these skills, and large companies are starting to treat them like internal playbooks for AI, he added.
Fortune 100 companies are using skills to “teach agents about their organizational best practices,” Murag said.
The rise of AI agents
Tech leaders have described AI agents as a potential game-changer for office work. OpenAI CEO Sam Altman said in June that AI agents are already performing tasks that are normally done by junior-level employees.
“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,” Altman said of AI agents at the Snowflake Summit 2025.
We’ll “start to see agents that can help us discover new knowledge, or can figure out solutions to business problems that are kind of very non-trivial,” Altman added.
Microsoft’s AI platform product lead Asha Sharma said in an episode of “Lenny’s Podcast” in August that AI agents could flatten corporate hierarchies.
“The whole kind of organizational construct might start to look different in a few years,” he said. “You just don’t need as many layers.”
But some people in the industry have said agents have been overhyped.
Guido Appenzeller, a partner at a16z, said on a company podcast episode in May that some startups are simply adding a chat interface to a language model and calling it an agent so they can charge more.
“A couple of startups are basically saying, ‘Hey, we can price this software that we’re building much, much higher because this is an agent,'” he said, adding that “there’s a marketing angle to agents.”
Read the original article on Business Insider
The post Anthropic researchers say the industry should stop building tons of AI agents — the real breakthrough is something simpler appeared first on Business Insider.
