AI Agent Startups Utilize Human Workers to Overcome Model Limitations
Despite advancements in foundation models from leaders like Meta and OpenAI, current AI agents struggle with complex real-world actions such as making phone calls or exercising common sense. To bridge this gap, new startups including Dazzle AI, Duckbill, and Wajo AI are implementing a 'human-in-the-loop' model where tasks that cause software to fail are outsourced to human contractors. Companies like Duckbill are providing APIs that allow AI developers to delegate difficult sub-tasks—such as scheduling appointments or negotiating bills—to people. This hybrid approach aims to ensure reliability and prevent the excessive 'token burn' associated with AI models attempting to solve tasks they are not yet capable of handling autonomously.
Tickers: $META
Topics: AI Agents Automation Challenges Foundation Models Human-in-the-loop