Pizza giveaway exposes AI agent growing pains

Driving the news: AI startup Hark offered $30 in DoorDash credit as part of a promotion designed to showcase its agent's ability to order food.

• The company advertised a $1 million giveaway, encouraging users to download Hark Pro, connect their accounts and ask the agent to order a pizza.
• The promotion helped propel Hark to No. 1 in the Lifestyle category of Apple's App Store, according to a screenshot shared by CEO Brett Adcock.
• Hark says it ultimately delivered nearly 20,000 pizzas, despite technical problems caused by overwhelming demand.

The catch: A number of customers reported experiences that were considerably less appetizing.

• Investor Sheel Mohnot said ordering took him 20 minutes, calling the promotion a failure even though he otherwise likes Hark.
• Another user reported that the agent placed an order despite the promotional credit not applying, resulting in a $27.70 charge to his brother's credit card.
• Replika founder Eugenia Kuyda posted screenshots of repeated unsuccessful attempts, joking that she'd never worked so hard for anything.

Zoom in: I repeatedly tried to get Hark to order my pizza, without success.

• I did receive the promised $30 DoorDash credit, but the agent couldn't complete the order.
• After connecting my Gmail account, I received a critical security alert from Google about an attempted sign-in. At least one other user reported a similar warning.

Zoom out: There's also a big privacy tradeoff. As this person highlighted, Hark requests extensive Google account permissions, including the ability to read, compose and send email.

• Such access can make AI agents more useful by allowing them to act on our behalf, but it also raises the stakes when users are being asked to grant broad permissions for something as mundane as ordering dinner.

What they're saying: Adcock acknowledged that the promotion strained Hark's systems.

• In an Oct. 17 post, he said usage surged to unprecedented levels and that many things broke.
• He said the company delivered thousands of pizzas end to end, including handling logins and payments, and ran more than 203,500 browser sessions through its Handoff model.
• A Hark spokesperson said the company ran into a bottleneck due to the massive demand, and the problems were unique to the promotion. The company offered gift cards to anyone who was charged or had a challenge redeeming the offer.
• "We're so happy the majority of people had a great night with a free pizza," the spokesman said. "Obviously, we hate to have let anyone down and we've made updates on how we handle peak surge requests."

The bigger picture: Hark's experience illustrates a challenge facing the broader AI agent industry.

• AI agents can navigate websites, fill out forms and complete purchases, but doing so often requires authentication, permissions and human intervention.
• Those extra steps can make an otherwise simple transaction more complicated.

Between the lines: Some tasks may be better suited to AI agents than others.

• In my recent testing of OpenAI's Dots and Meta's Muse, both proved useful at complex tasks like researching flights and planning events.
• But when I tried to have them buy basketball trading cards, Muse got stuck at Target's sign-in page. Dots simply directed me to Target's app, where I already had my payment information saved.
• Following those instructions myself turned out to be faster.

The bottom line: AI agents are getting better at doing things for us. But that doesn't necessarily mean they're making our lives easier.