Can AI find better deals on holiday purchases?

SEATTLE — The holidays are almost here, and whether you are looking to book a trip or find the perfect gift, everyone is trying to get the best deal possible.

Some are choosing to turn AI into their own personal shopper, so KIRO 7 put the tech to the test as experts weighed in.

Most internet users have now interacted with AI in some form. Whether through stills and videos generated and shared on social media, AI-based search results like Google Gemini or chatbots like ChatGPT, the tech seems unavoidable.

“Over the last few months or so, a lot of these companies that are known for chatbots have started saying, ‘Hey, it’s more than just a chatbot,’” Chirag Shah, a professor at the University of Washington’s Information School, said.

Shah’s been studying and writing about AI for years. He’s been tracking a shift away from the so-called “AI slop” made by generative AI into something else.

“Moving beyond the chatbots,” he said. “Moving from generative AI to agentic AI.”

Agentic AI is an artificial intelligence that can take action instead of just “thinking.”

“Across the board, there has been a 1,000 percent-plus increase year over year in people using ChatGPT, using Gemini, using Claude to find and research products,” Aniket Deosthali, the CEO of Envive AI, told KIRO 7.

Deosthali’s company creates AI chatbots to help online shoppers. His customers include the shapewear brand Spanx and the sunscreen brand Supergoop.

As a CEO, he’s a busy guy. Lately, he’s been delegating more and more of his tasks to agentic AI. He shared a recent example of having a bot coordinate between himself and a colleague to book a conference room for an upcoming business trip.

“It creates so much value to have that automation,” he said.

Several AI agents are available to the public, including Microsoft’s Copilot, Anthropic’s Claude and XAI’s Grok Bot.

In September, Meta launched Muse. The AI agent is free and, due to Meta’s ownership of Facebook and Instagram, many are encountering it. One of the biggest differences between Muse and the other agentic AI offerings is that users can watch Muse open a web browser and operate in real time.

KIRO 7 raced Muse against reporter Jack Bilyeu in three categories: finding the best deals on travel, toys, and tickets for the holiday season. Bilyeu was allowed five minutes to find the best deal in each category. The AI was then given as much time as it needed to find a better deal, but the clock continued to run until its responses were properly vetted.

The first challenge was finding a toy: a Needoh cube, the fidget toy that is constantly sold out.

After five minutes, Bilyeu found a questionable eBay listing for $7 or local pickup via Facebook Marketplace for $10.

Muse immediately ran into an issue. It repeatedly tried to access a shopping site, but was unable to get past the CAPTCHA proving it was a human. It then pivoted and found an authentic Needoh available for in-store pickup at a Designer Shoe Warehouse location in Redmond. It took five minutes and 21 seconds.

The next challenge was finding the cheapest flights from Seattle and Honolulu, leaving December 23 and returning December 26; two of the busiest travel days of the year.

The cheapest tickets Bilyeu found in five minutes cost $1,854.

Muse took 29 minutes and 30 seconds to find a trip for $1,684. The savings came from booking through two different airlines. It took so long because it repeatedly stopped to ask Bilyeu for more details.

The final test was booking two tickets to either of Olivia Rodrigo’s sold-out concerts at Climate Pledge Arena. Within five minutes, Bilyeu found a pair of tickets for $604 on Gametime.com. After six minutes and 47 seconds, Muse found the same deal on the same site.

What did these tests prove? Depending on what you’re after, your mileage assigning your holiday shopping list to an AI agent may vary.

Bilyeu was on an artificial time crunch, and the AI was not utilized to its fullest extent. For example, it could have been assigned to search for the best deal over the next 24 hours. Giving it credit card and location information also could have empowered the bot to make the purchases immediately, though several financial institutions, including Bank of America, have warned against it.

“I don’t want anyone sharing personal information with their chatbot,” Jill Schlesinger, CBS News business analyst, said. “These chatbots are not actually regulated in the way, say, your financial services companies are, so this information can float out there.”

The tech is not perfect, but it is still new and growing. Experts estimate it will still be anywhere from a few months to a few years before agentic AI is used as frequently as regular chatbots.

“When are grandma and grandpa going to be using this?” Bilyeu asked.

“Not today, but the technology is maturing so quickly,” Shah said.

Watch KIRO 7 News at 5:30 p.m. on Friday for the full story.