AI agents continue to rely on human labor for complex tasks
Many consumer AI startups are incorporating human contractors into their workflows to complete tasks that current foundation models cannot reliably handle.
While companies like Meta and OpenAI advance the capabilities of AI agents to browse the web, the reality of deploying these tools often involves human intervention. Startups such as Dazzle AI and Duckbill are utilizing human workers to handle sensitive or complex tasks, including phone negotiations and scheduling, where current models frequently struggle.
Marissa Mayer, CEO of Dazzle AI, noted that while frontier models are effective at generating candidate lists and suggestions, they lack the common sense or execution reliability required for real-world services. The human element ensures that the service delivered meets quality standards, effectively acting as a quality control layer for automated suggestions.
Duckbill has launched an API specifically designed to allow other AI-agent companies to delegate tasks that their software is likely to mishandle. CEO Meghan Joyce stated that the company breaks down complex requests into granular, bite-sized pieces that human contractors can complete, mimicking the operational model of gig-economy platforms.
This trend highlights the current limitations of large language models in navigating the physical world. For the foreseeable future, companies are positioning human oversight as an essential component of their agent technology to bridge the gap between automated suggestions and successful outcomes.