Key points. * Marissa Mayer's Dazzle hands some agent tasks to people * Duckbill sells human assistance to other agent makers * Wajo brings in people for calls and CAPTCHA tests Some AI agent startups are sending tasks their software can't reliably finish, such as phone calls and bookings, to human workers. "We're not even trying to hide it, we think the human in the loop is a benefit," said Marissa Mayer, the former Yahoo CEO who now runs Dazzle AI. Duckbill and Wajo AI also bring in people when their agents fall short. Duckbill has started selling that human assistance to other companies building agents, according to CNBC on October 31. A similar setup drew privacy concerns inside Meta Platforms (META) in September. Dazzle has people step in when the agent gets confused. Mayer said today's top AI models are good at finding home renovation services but might fail when they need to book and schedule a rooftop upgrade. Dazzle sends certain tasks to human contractors. The company relies on the frontier models "to give us good suggestions and a good candidate list, but we're checking that over," she said. One Dazzle user asked the agent to handle a roof replacement, a broken window, and the purchase of a glass tabletop at the same time. The agent confused window glass with tabletop glass and ended up "trying to figure out if he could make the tabletop fit in the window," Mayer said. Dazzle's system flagged the task and passed it to a person. "Within a few seconds, the human had things sorted out. Whereas an agent would have burned a ridiculous number of tokens," Mayer said. Dazzle raised $8 million in December 2025 in a round led by venture capitalist Kirsten Green that valued the company at $35 million. Duckbill sells human assistance to other agent makers. Duckbill CEO Meghan Joyce said agents often get stuck on tasks like making a phone call, scheduling a doctor's appointment, or booking a local cleaning service. For tasks that need real-world judgment, she said, humans still beat AI. Duckbill recently launched an application programming interface, or API, that lets companies building personal agents hand those tasks to human contract workers. An API is a set of rules that lets one piece of software send requests to another. "We work on a model that is not dissimilar from Uber, where we boil tasks down into atomic bite-sized pieces," said Joyce, who previously worked at Uber (UBER). Duckbill's developer page lists a price of $5 per resolution and says the first 1,000 API calls are free. Its setup guide for AI assistants puts it another way: each phone call costs about $5, or two credits, and the first 1,000 calls are free during a launch period. Neither page explains how a resolution relates to an API call. The developer page says Duckbill has more than 200 human operators, all of them background-checked. Duckbill says its service can raise an agent's task completion rate from 60% to 95%. The page doesn't explain how those figures were measured. Wajo uses people for calls and CAPTCHA tests. Shivani Poddar, CEO of Wajo AI, is using human workers to handle phone calls and even CAPTCHA tests. Those are the puzzles websites use to keep bots out. "I think AI plus humans, collaborative AI, is what you need to actually get the job done," Poddar said. She said people will stop using agents that can't complete tasks. Wajo launched its agent, Fo, in late September. Fo can call businesses to book or cancel services, and it can hire a human to complete a task when needed. Poddar previously worked in engineering roles at Meta, Google, and DeepMind. Meta tested human callers for its Muse agent. Meta employees raised privacy concerns about an internal test that used human contractors to make phone calls for the company's Muse AI agent. A Meta spokesperson said at the time that this kind of testing is "core to the product development process." In that test, a Meta vice president said human callers could raise the success rate to 95% to 98%, compared with a lower rate for AI calling. As of October 31, Meta wasn't using humans to complete users' Muse requests. Duckbill and Dazzle say they screen their workers. Joyce said Duckbill does "full background checking" on the 1099 contractors it uses. A 1099 contractor is an independent worker, named for the IRS form used to report that person's pay. She compared them to drivers for Uber and DoorDash (DASH), "people who live here in the United States and want a source of additional income." Duckbill also uses contract workers from businesses that typically supply contact center labor to companies. Mayer said Dazzle runs background checks and screens for "people who we think are extremely discreet and have good taste and good judgment." She called the current stretch a "transitionary period" that will last until AI models can handle more involved web tasks and phone calls as efficiently as people. The AIStockWire daily brief Top movers, notable insider buys and fresh Congress trades in your inbox each morning. Free. No spam, ever. Unsubscribe in one click.

Tempo has raised $4.5 million to enhance its personal assistant service on WhatsApp, which uses AI for initial interactions and human support for complex tasks. Founded by Eduardo Latache and Raul Lima, Tempo targets high-income clients, offering services like bill payments and travel bookings. The company plans to expand its B2B offerings and develop new service packages. The funding round included investments from Caravela Capital and other notable investors.