In just five months, six new always-on AI agent services reached consumers. Google's Gemini Spark arrived in May, followed by xAI's Grok Bot and the invite-only Instinct in August. Meta launched Muse on September 8, OpenAI shipped Dots three weeks later, and Anthropic opened a waitlist for Claude Duet on October 21. Each one runs on its own cloud computer, remembers what its owner likes, and can browse, text, call, and purchase on that person’s behalf. Unlike the chatbots before them, they do not wait to be asked.
Muse shows how fast this is moving. On Meta’s Q3 2026 earnings call, Mark Zuckerberg said that “more than 10 million people in the US and Canada have set up Muse personal agents” in a month and a half, and that about 25% of them use it on any given day. Shopping is the most common use. More than a quarter of Muse users shop with it, he said, “with an average purchase size over $200.” In the first week, Meta had to ask Stripe to raise credit limits because volumes “were above what any of us expected.”
That $200 figure will put a sparkle in every marketer’s eyes, but it does require a caveat. When an analyst asked whether it covered only retail, Zuckerberg said it “does include travel, too.” A few flights and hotel stays pull an average up fast. Meta has not shared a median order or a category breakdown, so nobody outside Meta knows what a typical Muse basket looks like yet.
What Meta wants is clearer. Zuckerberg described a near future in which buying something works differently: “If you want to buy something, it’s not just navigating to that business’ website. It’s talking to their agent.”
He also said Meta expects to build a model where it gets “a percent of the value” it helps create. For digital marketing leaders, that changes the brief. The customer journey now runs from a person, to their agent, to your brand, and back to a person who lives with the result. The practical question is which of your experiences that agent will reach first.
Consumers want to avoid annoying interactions with your brand
That may be a hard thing for marketers to hear, but many brand interactions aren’t as smooth as consumers would like, and they’re willing to throw their agents at those interactions first. The American Customer Satisfaction Index (ACSI) scores US industries each year. Scores range from 0 to 100, with 100 being the best score. Its 2026 studies put full-service restaurants at 82 and specialty retail at 80. Internet providers sit at 73 and subscription TV at 72. Then look at what consumers want an agent to do. A Lorikeet survey in October asked 1,966 US adults how comfortable they would be with an AI agent contacting a company on their behalf. The most popular task was getting a better price, which 55.5% of consumers were comfortable delegating. Making a complaint came next at 41.8%, followed by making a booking at 38.5%.
Put the two side by side. Haggling over price and complaining are the conversations people least want to have, and they cluster in the industries that score lowest: the retention call, the billing dispute, the claim, the cancellation. That's is where we expect agents to show up first. But it’s still early. In the same survey, only about 6% of consumers said an AI had already contacted a company for them, while 66% were interested. Gartner found that 51% of customers would use an AI assistant for service on their behalf, and warned that this “poses a cost risk that could undermine any gains made by customer service automation.”
What changes, industry by industry
Telecom and Internet
In ACSI’s 2026 entertainment study, call-center satisfaction is subscription TV’s lowest-rated benchmark, at 72. Meta says Muse can lower bills and negotiate for its owner, and Instinct now calls businesses to get things done. Picture a retention desk built on hold times and save offers, facing a caller that never tires, calls back every few hours, and knows every competitor’s price. Verizon is already moving in a different direction. Its Q3 results show promotional retention costs down 30%, which its CEO credited to “[Verizon’s plan called] Simplicity and our expanded loyalty program over heavy subsidies.”
Subscriptions and Streaming
About 40% of consumers cancel at least one paid streaming service a year, and 61% would drop their favorite over a $5 monthly increase, according to Deloitte research cited by ACSI. Rocket Money’s Rowan agent, built with Anthropic, texts users before a free trial converts and cancels the service if they reply “cancel.” Revenue that depends on customers forgetting will shrink.
Insurance
Consumers will let agents shop before they let them act. In an Insurity survey, 46% would let AI generate a quote, but only 22% would let it file a claim and 16% would let it cancel or renew a policy. ACSI’s Forrest Morgeson notes that insurer AI still “sits behind the scenes in claims and underwriting” Expect agents to show up first at renewal, comparing quotes in bulk.
Travel and Hospitality
Travelers are wary of letting an agent book. In an Expedia and YouGov survey of more than 5,700 adults, 53% would let AI suggest options, but only 8% are comfortable booking through an AI platform. The supply side is moving faster anyway. Google’s AI Mode now books hotels with Marriott, Hilton, IHG, and others acting as merchant of record, and Hilton Honors points work inside it. Hilton CEO Chris Nassetta told investors to “assume we’re working with each and every one of [AI agents].”
Retail
Agents compress comparison shopping into seconds, and loyalty is thin. Checkout.com found that 57% of consumers would let an AI shopping agent switch brands if it found a better value. Retailers are split in their response. Amazon blocked Muse 12 days after launch. Walmart opened an Agent Gateway on October 21, with profiles that let customers approve each purchase or set a spending cap. Walmart also learned that when checkout happened inside ChatGPT, it converted three times worse than sending shoppers to Walmart.com. Shoppers who use Sparky, Walmart’s own agent that’s also inside ChatGPT and Gemini, spend about 35% more per order.
Agents also favor what they can buy. A Fast Company test found that every product Muse recommended had an active checkout page.
Home Goods
Big, heavily considered purchases move slowly. Wayfair CEO Niraj Shah says that with agent traffic today, “you have to put a high percentage on a very low number,” and that the hard part of home is customer service and logistics. ACSI found the ease of arranging a repair fell 7% for TVs and PCs this year. Home Depot has put conversational AI across text, chat, and phone and plans to join agentic shopping in Google’s AI Mode and Gemini. About six in ten merchants surveyed by PYMNTS would keep big-ticket items away from agents for now.
Restaurants
Dining scores highest on satisfaction, but delivery sits at 75, and the fairness of fees is its weakest benchmark at 73. Restaurants want agents on their own terms. Resy banned accounts after one Instinct agent pinged it “hundreds of times every hour” for a table, while text agents such as Zest book through OpenTable and Resy directly. DoorDash CEO Tony Xu calls agents “new forms of the Googles,” and argues that customers return to whoever fixes the late or missing order.
It’s a convenience today, a strategy problem next
The volume is still small. Similarweb finds AI refers just 0.4% of retail ecommerce visits. More than two in five consumers would stop using an agent for a task after one mistake, according to S&P Global. But the shoppers who arrive through AI now behave better than everyone else.
Forecasts for where agent usage will land vary widely, mostly because each firm defines the market differently. Morgan Stanley expects agents to drive $190 billion to $385 billion of US ecommerce by 2030. Juniper Research projects $1.5 trillion in global agentic spending that year. eMarketer puts direct sales on AI platforms at about $20.6 billion this year, roughly 1.5% of US retail ecommerce.
Our read is that agents are a convenience layer in 2026: faster research, fewer errands, a human still approving the purchase. The long-term contest is over who owns the customer’s identity, the moment of decision, and the data from every interaction. Brands that wait until agents carry real volume will negotiate from a weaker position.
Provocations for the next 18 months
- Assume your retention desk will negotiate with software. Telecom, streaming, and insurance brands should publish retention offers and cancellation terms as rules an agent can read, and decide what a customer is worth before the agent asks.
- Stop counting on friction. If revenue depends on hold times, buried cancel links, or forgotten trials, an agent will find it. Replace inertia with value the customer notices every month.
- Own what happens after the sale. An agent can book the hotel or order the sofa. It can’t redeliver a damaged sofa or rebook a canceled flight without you. Make delivery dates, rebooking rules, and repair scheduling readable and actionable.
- Make loyalty visible to the agent. Hilton Honors already works inside AI Mode. As Merkle has written, “the loyalty programs that thrive in agentic commerce won’t be the ones with the most points. They’ll be the ones agents trust, recognize, and recommend.” Start measuring how often agents recommend you.
- Build one front door for every agent. Hilton is working with every agent, and Walmart built a single gateway. Meta has said it wants a percentage of the value it creates, so every one-off deal can become another fee.
- Price as if every fee will be compared. Delivery fees already score lowest with customers, and the FTC has proposed that personalized prices be disclosed along with the data behind them.
- Learn to see agent traffic. In the PYMNTS survey, no merchant could consistently trace AI visits to a completed sale, and 12% could not tell agents from humans at all. HUMAN Security says Muse alone made 72% of the agent requests it observed in early data.
On timing, start with visibility this quarter. Adobe expects AI traffic to US retail sites to grow 130% this holiday season, making it the first holiday when agents matter. Within six months, stand up an agent gateway and link loyalty identity to it. Within a year to eighteen months, put your own agent inside the platforms where customers already are.
This work crosses data, loyalty, commerce platforms, and the contact center, which is why it rarely fits inside one team. It is the kind of end-to-end change Merkle helps clients plan and deliver. A good first question for your team: if a customer’s agent asked for your best offer at 2 a.m. tonight, what would it get back?
We would like to hear how this looks from where you sit. Which of your customer conversations do you expect agents to reach first, and how ready is your brand to answer them? If you want to compare notes, or you’re ready to set your brand up for the customers who arrive by agent, get in touch with Merkle. We’ll start with what those agents see today and build from there.
