--- title: "How AI agents could change restaurant ordering" url: https://burnaze.com/article/ad30af0d-0728-431c-8600-eec10c1e753e source: "Restaurant Business" original_url: https://restaurantbusinessonline.com/technology/how-ai-agents-could-change-restaurant-ordering category: "Pizza" industry: "Restaurants" segment: "Pizza" topics: ["Technology"] published: 2026-10-13T02:32:15.000Z modified: 2026-10-31T14:58:10.119Z image: https://cdn.informaconnect.com/platform/files/public/2026-10/background/shutterstock_1672275235.jpg?VersionId=ax7djR0roqgti1p_HZmEd2KhDOrLYaIN --- # How AI agents could change restaurant ordering **TL;DR:** AI agents, such as Muse, can order food from restaurants like Pizza Hut, potentially changing the way customers interact with restaurant websites and apps. This new technology may threaten third-party delivery marketplaces like DoorDash and Uber Eats, as it allows customers to order directly from restaurants without marked-up prices or commissions. *originally published in Restaurant Business on 2026-10-13. [Read the original](https://restaurantbusinessonline.com/technology/how-ai-agents-could-change-restaurant-ordering).* ## Key facts - AI agents can order food without special integrations - Restaurant websites may become less relevant - AI agents could reduce third-party delivery markups ## Story In 1994, Pizza Hut quietly launched what may have been the first online ordering site. Called PizzaNet, the bare-bones portal allowed customers in certain areas of Santa Cruz to view a rudimentary menu, type in their order, address, and phone number, and have the info sent to the nearest Pizza Hut, via the chain’s headquarters in Wichita, Kansas, thanks to a new technology called the World Wide Web. An employee at the store would then call the customer to verify the order. Payment was handled at the point of delivery. A Los Angeles Times article described the novel ordering system as “half-baked,” noting that while it had potential, it was still less efficient than calling the store directly. ## Early online ordering But Pizza Hut correctly foresaw that the internet would one day be a legitimate ordering channel for restaurants. In the decades to come, the web transformed how restaurants and retailers do business by giving consumers the ability to buy things from the comfort of their homes. Today, restaurants’ websites and mobile apps are essentially second storefronts, with sophisticated menu designs and marketing tools designed to get customers to spend more and come back more often. The internet has also given rise to delivery networks like DoorDash and Uber Eats that have made delivery an option for virtually any restaurant. While these innovations have come with their share of headaches, they have also allowed restaurants to reach more customers than ever before. And now artificial intelligence is threatening to upend all of it. The watershed technology has already changed quite a bit about how people use the internet. Now when you Google something, the top result is often not a website link, but an AI-generated summary that may not even require a click. In the AI era, you can surf the web without even using a search engine: Just type your query into a chatbot like ChatGPT or Google’s Gemini and let them do the searching for you. These developments are impacting restaurants to varying degrees. If you ask ChatGPT for a couple of restaurant recommendations nearby, for instance, it’s probably going to provide different results than if you’d typed the same thing into Google. ## AI agents emerge But an ever bigger disruptor to restaurants’ digital business could be AI agents, the latest generation of AI bots that can not only respond to queries but also take action on users’ behalf, including placing orders from restaurants. As recently as late last year, shopping agents were mostly theoretical. But over the past few months, a number of them have hit the market, including the invite-only Instinct, OpenAI’s Dots, and Muse, from Facebook parent Meta. These agents are designed to perform a wide range of tasks when given access to the user's personal accounts like email and banking. For instance, while playing around with Muse today, it alerted me that an Amazon package had been delivered, which it knew because I allowed it to read my emails. I also used Muse to order a pizza from Pizza Hut, and the process illustrated some of the opportunities and limitations of this emerging ordering mode. Starting a restaurant order with an AI agent really could not be easier. You can just type “pizza” and the bot will get the idea and begin narrowing down the details, like where and what you’d like to order. Once it has that info, it does the rest, and it doesn’t need any special integrations with Pizza Hut to do it. Muse has its own web browser and navigates the Pizza Hut site by clicking and scrolling around, just like a human would. It creates an order and presents it for review. Payment can be processed directly within Muse through a secure service called Link. You can also have Muse take you to the restaurant’s website to complete the order there. This seemingly simple process is chock-full of implications for restaurants. Maybe most obviously, it makes the restaurant’s website as currently constructed virtually irrelevant to the customer. Because the agent is doing all the work, I don’t have to go to pizzahut.com, which means I don’t see that the chain is currently offering a $7 meal deal promotion; I don’t read the menu descriptions or see the photos; and I’m not hit with an upsell tempting me to add a side of wings or breadsticks. In an agentic future, restaurant websites may act less like billboards and more like literal menus. They’ll be optimized for AI agents, which favor straightforward data, unlike humans, who can be easily manipulated by pictures of delicious food. This type of optimization is already happening elsewhere on the web. The news outlet Time, for instance, has been converting all of its articles from HTML to a format called markdown that is easier for AI bots to read. (It’s also reportedly selling ads on these markdown pages designed to influence AI bots.) Restaurants will probably end up having to do something similar. Indeed, many have already updated their websites to be more AI-friendly, which has led to the sudden ubiquity of FAQ pages — just the sort of structured information that large language models look for. But I doubt agentic ordering will cause restaurant websites or apps to go away entirely, just like restaurant websites haven’t replaced brick-and-mortar restaurants. There will always be customers who want to use these “old-fashioned” channels, and restaurants can’t afford to lose customers. Plus, web or app ordering will likely remain the preferred channel for restaurants, as they will continue to be the truest digital expression of the brand. ## Implications for restaurants That was one of several things that jumped out to me about the agentic ordering process: **No markups.** When I ask Muse to order from Pizza Hut, it goes directly to the website of the nearest Pizza Hut rather than a third-party app. This is good news for both me and Pizza Hut, because it means I don’t have to pay marked-up prices, and Pizza Hut doesn’t have to pay hefty third-party delivery commissions. It also gets to keep whatever personal information I share — in this case, my address, email address, and phone number. Restaurants have been trying to get customers to order directly from them instead of through DoorDash or Uber Eats for years, and AI agents present a potential workaround. Some observers have seen enough to declare agentic ordering a real threat to third-party delivery marketplaces. When asked about that possibility earlier this year, DoorDash CEO Tony Xu argued that delivery apps will continue to thrive because they handle the entire ordering process from start to finish, which provides a better overall experience for the customer. Meanwhile, DoorDash has been adding more agentic features into its own app, including a chatbot that can help customers search for restaurants as well as the ability to order from DoorDash via text message. **Loyalty is a pain point.** Muse did not ask if I wanted to sign in to a Pizza Hut loyalty account. If I wanted to get loyalty points for my Pizza Hut order, I’d have to ask Muse to sign in, or sign up, for me. Then I’d need to share my Pizza Hut login information with Muse, which requires me to fill out a separate form for secure credentials. As a general rule of thumb, the less friction in a digital transaction, the better, and there’s a lot of friction here. This is a technical problem restaurants may need to be proactive about solving if they want loyalty to be part of agentic ordering. **The use case is narrow.** An AI agent works well if you know exactly what you want. But if you’re ordering from somewhere new, or don’t know what you’re craving, the benefit is less obvious. You *could* ask the agent to make a suggestion, based on what it knows about you. But control freaks, or people who just enjoy browsing, may not be into that. And it’s not necessarily more convenient, anyway. Speaking of which… **Muse is slow as molasses.** It took Muse four and a half minutes to create an order for one pepperoni pizza from Pizza Hut, a task that I was able to accomplish myself in about 40 seconds (not to brag). Meta’s excuse would probably be that Muse is meant to “work in the background” while I do other things. But I struggle to see how that is going to radically boost anyone’s productivity. Ultimately, agentic ordering’s impact on restaurants will come down to how many consumers choose to adopt it. And right now, adoption is still quite low. In a recent post, venture capital firm Andreessen Horowitz (a16z) wrote that early AI agent users tend to already be power users — the miniscule fraction of consumers who spend hundreds of dollars a month on AI subscriptions. These “prosumers,” as a16z calls them, are apparently not really using agents to order food. In an analysis of more than 1,500 power user group chats, the most-discussed topic was coding and automation, with 103 mentions. Dining and reservations was the least-discussed topic, with just 16 mentions, a16z found. To put this into further perspective, downloads of Muse during the first three weeks after it launched paled in comparison to downloads of Threads, Meta’s answer to Twitter. Muse had about 5.4 million downloads over that time, while Threads had 15.7 million, showing that consumer interest in agents is still relatively low. “As with every new consumer product, the primary battle is against apathy,” a16z wrote. Of course, we’re still in the very early days of so-called agentic commerce. And there are some who believe it could become widespread fairly soon. A recent report from Mastercard, for instance, predicted that 1 in 10 people will be using AI agents to shop and pay by 2030. It found that young people in particular are likely to be eager adopters of AI agents, noting that 27% of teens said they would use AI to help them shop, compared to 16% of parents. It’s also worth remembering that PizzaNet, too, seemed a bit unnecessary back in 1994. As the LA Times noted, there wasn’t much difference between ordering a pizza over the phone and online. “The transaction is basically a commodity,” author Michael Schrage wrote. “The trick is to figure out new ways to create value for the customer.” AI agents have a few ways of creating that value, namely, by saving consumers time and money. But they have some bugs to work out, too. If they can do that, they have a chance to make their own mark on restaurant ordering. ## Entities - QSR (sector) - Restaurant Business (publication) - Pizza Hut (company) - DoorDash (company) - Uber Eats (company) - digital transformation (topic) - third-party delivery (topic) - ai agents (topic) - Tony Xu (person) - Muse (company) - restaurant ordering (topic) - Muse investment in Pizza Hut (deal) --- Originally reported by Restaurant Business (https://restaurantbusinessonline.com/technology/how-ai-agents-could-change-restaurant-ordering). Summarized by Burnaze (https://burnaze.com/article/ad30af0d-0728-431c-8600-eec10c1e753e).