Autonomous AI agents still struggle with real-world tasks like phone calls and scheduling, leading startups such as Dazzle AI, Duckbill, and Wajo AI to quietly route complex chores to human contractors behind the scenes.
The current race to deploy autonomous digital assistants has run into an awkward bottleneck: the software simply cannot reliably handle messy physical-world errands. While frontier models from companies like Meta and OpenAI excel at surfacing online shopping options or summarizing web pages, they routinely stumble when tasked with making phone calls, reserving local cleaning services, or coordinating complex appointment schedules, according to a CNBC investigation. Behind the polished veneer of automated software, a growing number of startups are relying on human operators to bridge the gap between AI promises and everyday reality.
How Dazzle AI, Duckbill, and Wajo AI Route Tasks to Human Contractors
Rather than letting autonomous systems fail in public, developers of personal agent platforms are building human safety nets directly into their workflows. Startups including Dazzle AI, Duckbill, and Wajo AI are quietly deploying human workers to handle phone calls, manage edge cases, and solve CAPTCHA tests designed to block bots.
Duckbill recently introduced an application programming interface specifically designed to connect personal agent software with human contract workers. Marissa Mayer, who now leads Dazzle AI, explained that her company reviews suggestions generated by frontier models before executing them to ensure high service standards.
We’re relying on the frontier models to give us good suggestions and a good candidate list, but we’re checking that over to make sure that we’re really delivering the type of service that we can stand behind.
Marissa Mayer, via CNBC
Why AI Models Confuse Glass Tabletops with House Windows
The underlying models powering today’s digital agents frequently lack basic common sense when executing multi-step physical projects. Mayer shared an example involving a Dazzle AI super user who asked the technology to simultaneously coordinate a roof replacement, repair a broken window, and purchase a glass tabletop.
The AI agent confused window glass with tabletop glass, attempting to figure out if he could make the tabletop fit in the window.
Dazzle’s monitoring technology quickly flagged the error and handed the job to a human operator.
Within a few seconds, the human had things sorted out. Whereas an agent would have burned a ridiculous number of tokens. Or they may never have gotten to a good solution.
Marissa Mayer, via CNBC
Meanwhile, Wajo AI CEO Shivani Poddar emphasized that collaborative AI combining software capabilities with human judgment is essential to prevent user abandonment. However, this hybrid reliance raises thorny questions regarding transparency, pricing, and data privacy when consumers pay premium rates for automated services delivered largely by human labor.
Background Checks and Privacy Concerns in the 1099 Workforce
Managing a hidden human workforce introduces operational challenges familiar to the gig economy. Duckbill CEO Meghan Joyce noted that her startup draws parallels to ride-hailing platforms like Uber, breaking down complicated requests into atomic, bite-sized pieces for human contractors.
Duckbill executes full background checking
on its 1099 independent contractors, drawing workers from the United States seeking supplementary income as well as established contact center businesses. Dazzle AI similarly enforces vetting standards modeled on traditional personal assistants.
We generally are screening for people who we think are extremely discreet and have good taste and good judgment, both in terms of how they interact with our users, but just the ability to kind of forget and compartmentalize that information afterwards.
Marissa Mayer, via CNBC
Privacy remains a central concern for users interacting with personal agents. Meta previously faced internal employee pushback over testing human contractors to handle phone calls for its Muse AI agent. While Meta currently does not use human operators to complete public requests, a spokesperson defended testing as core to the product development process.
Mayer acknowledged the industry is navigating a transitionary period, noting that Dazzle makes no secret of its human-in-the-loop architecture.