“AI can do 80% of an employee’s work,” startup founder says, raising staffing fears

A South Korean startup founder says an AI agent he built himself is performing at roughly 80% of one employee’s level, leaving him questioning how his company should organize its workforce.

The founder, identified only as A, said in a post last month that he runs an IT startup with about 15 employees. He had given staff unrestricted access to various paid AI models and covered the costs, but said the resulting improvement in practical productivity remained far more limited than expected after several months.

“

Honestly, its performance has reached about 80% of one employee.

The outcome changed when A, who has an engineering background, built an AI agent and assigned it work. Unlike a tool that simply follows a person’s instructions, an AI agent can handle multiple tasks on its own. “Honestly, its performance has reached about 80% of one employee,” he wrote. He added that feedback was easier because he did not need to worry about the agent’s feelings and communication was more convenient. From a strictly business perspective, he said, he had begun to think that replacing employees with agents might be the wiser choice.

“

I even reached the point of thinking that, strictly from a business operations perspective, getting rid of employees and using agents might be the much wiser choice.

A said he did not actually want to fire his staff. But he also wrote that only a small number of highly advanced AI users might survive in the emerging structure, and that leaving the gap between labor costs and productivity unaddressed could amount to failing in his duty as a CEO. He said he wanted to improve the team’s abilities so everyone could grow together, while acknowledging that people who do not take initiative with AI are quickly falling behind in both productivity and understanding.

Commenters said companies should do more than provide AI tools. Managers, they argued, need to teach employees how to use them and embed them in daily work. Others compared AI adoption to providing Excel, saying access alone does not produce equal results, and said their companies train workers in prompt writing. Some commenters predicted restructuring, while A later replied that a one-person company might feel easiest in the age of AI.

“

In the end, I think a structure will emerge in which only a small number of highly advanced AI users survive, but as a CEO, I sometimes feel that leaving this gap between labor costs and productivity unattended would be a dereliction of duty.