The sudden surge of strange phone calls hitting small businesses across the country has become a talking point in tech circles and local newsrooms alike. Callers introduce themselves with a friendly tone, disclose that they are transcribing the conversation, and then dive straight into very specific requestsโwhether itโs a quote for installing three bidets, a reservation for a karaoke room, or a sameโday cake pickup. Some recipients go along with the flow, answering questions and logging the order, while others pause, suspecting that the voice on the line is not human. A Seattle karaoke clerk bluntly asked, โAre you AI?โ and insisted on speaking with a real person, while a Colorado plumber, bewildered by the request for a โcustomerโsuppliedโ bidet, fired back with a vulgar retort that was automatically transcribed in a garbled fashion. These encounters are not isolated pranks; they stem from a new wave of AIโdriven assistants that are being let loose on the public telephone network, forcing Main Street merchants to grapple with a technology that blurs the line between helpful automation and intrusive spam.
At the heart of this phenomenon is Muse, an AIโassistant tool unveiled by Meta last month and marketed as a โhandleโitโallโ companion capable of online shopping, flight purchases, restaurant reservations, inbox management, and, for a subset of users, placing phone calls on behalf of its human operators. Internal documents leaked to 404 Media suggest that some of these calls may actually be hybrid operations, where a callโcenter agent takes over after the AI has gathered basic information and then hands the task back to the machine. From the perspective of the smallโbusiness owner, however, the distinction matters little: the interaction feels like speaking to a bot, and the traditional expectation of a human counterpart is disrupted. Early adopters on platforms such as X have shared mixed reviews, with one AIโfocused YouTuber noting that the system works impressively when it succeeds, yet roughly half of the businesses contacted simply hang up on the agent. This split reaction underscores the growing tension between the promise of seamless automation and the reality of consumer resistance.
To understand the human side of this equation, I reached out directly to the workers who have been on the receiving end of these calls. The Colorado plumber, after a brief laugh, hung up again, clearly uninterested in engaging further. The woman at the Seattle karaoke venue was more measured, expressing confusion about whether she was speaking to an AI, a salesperson, or a genuine customer. After clarifying that I was a human researcher interested in her experience, she confirmed that her establishment actively dislikes AIโinitiated calls and prefers to deal with real people. Her testimony highlights a broader sentiment: many service providers view unsolicited AI outreach as an invasion of their workflow, an unwanted addition to the already busy task of managing customer inquiries. Even when the AI manages to secure a bookingโas happened when a subsequent agentic tool succeeded after Muse failedโthe underlying discomfort remains, suggesting that mere functional success does not equate to acceptance.
Beyond the quirky anecdotes, these interactions signal a deeper shift in how smallโbusiness operators are being compelled to adapt. In sectors where owners traditionally field calls or emails, negotiate prices, and then deliver a service, the arrival of AIโdriven callers feels novel, awkward, and at times antisocial. If the tech industryโs vision prevails, such encounters could become far more commonplace. Museโs rapid adoptionโmillions of downloads shortly after launchโwas preceded by the rise of Instinct, an inviteโonly personal assistant valued at around $10โฏbillion that had already captured attention in the Bay Area as a โtrained to use a phone and a computerโ agent. A few weeks later, OpenAI unveiled its own offering, dubbed Dots, described as โalwaysโon agents built to handle everything.โ While booking a plumber for a bidet might seem like an edge case, businesses that are already heavily onlineโurban restaurants, concert venues, and techโcentric recreation facilitiesโhave been contending with rudimentary bots for years and are now bracing for a more sophisticated wave.
The overarching promise is simple and familiar: every individual will possess a tireless automated assistant working on their behalf, and tech conglomerates are earmarking hundreds of billions of dollars to see if this dream can finally be realized. In the near future, millions or even billions of agents could be online, executing instructions by moving cursors, clicking buttons, typing requests, probing every humanโfacing interface for an edge, and occasionally picking up the phone to ask, โWhat can you do for us?โ Even before Muse began placing calls, nonconsensual AI interactions had become routine. Chatbots populate virtually every text box, productivity apps glitter with magic buttons and sparkle icons, and the digital landscape is saturated with AIโgenerated contentโendless streams of text and images that have clogged social feeds, music platforms, eโbook stores, comment sections, job boards, and public submission forms. This deluge of synthetic output has strained many online systems, and if it serves as a preview of a world where Muse manages all customerโservice calls, the picture is far from inspiring.
Earlier in the year, a group of researchers examined 84 suspected spikes in AIโdriven usage of online government servicesโranging from freedomโofโinformation requests to publicโcomment processes, benefit enrollments, and fine appealsโacross the globe. They coined the term โagentic floodingโ to describe what occurs when the deโfacto rate limits that once kept demand in check disappear. Lead researcher Chris Schmitz explained that in scenarios where administrative burdens prevent citizens from accessing services or where people are simply unaware of available benefits, AI could be a powerful force for good. Conversely, when automated submissions overwhelm a publicโcomment channel, the same technology can trigger a tragedy of the commons, degrading the quality of discourse and hindering legitimate participation. Prior to Museโs launch, even primitive agentic tools were already creating strain, prompting governments to consider two divergent strategies: expanding capacity to meet the new demandโthrough hiring more staff or deploying their own AIโor introducing additional friction to suppress the influx. Schmitz warned that building friction is cheap, precedented, and politically attractive, raising concerns that authorities might choose the easier, restrictive path rather than investing in genuine accommodations.
A few weeks after Schmitzโs paper entered the public discourse, Meta released Muse, and one of its highlighted use cases was directing the agent to state unclaimedโproperty websites, where it would autonomously open a browser, navigate, scroll, and fill in personal data. My initial attempt to use this feature faltered when the NewโฏYork State Comptrollerโs site rejected Muse outright, flagging it as a bot. On a second try, I assisted the agent with manual mouse movements, after which it surfaced three potential claims and offered to file them immediately. The path of least resistance would have been to accept the agentโs suggestion, but upon closer inspection I discovered that one of the claimsโa modest payment from a media companyโwas incorrectly valued in the thousands, while the other two clearly belonged to other individuals. I filed the legitimate claim by hand; Muse had been prepared to submit all three, leaving the state to sort out the rejections. This episode illustrates a core tension: the agent possesses virtually unlimited time and computational resources, so it has little incentive to conserve effort or avoid duplicative work, even when doing so creates unnecessary burdens for downstream human administrators.
Proponents of Muse and Instinct often celebrate the liberation from drudgery, claiming that these virtual assistants finally perform tasks that once required a human touch. Companies envision extending agentic logic into chat environments, suggesting that group conversations could be streamlined by an AI participant that coordinates schedules and makes plans without the usual backโandโforth. A recent tweet demonstrated Instinct joining a group chat and promising to handle logistics for the whole crew, accompanied by a link to a short video clip. There is undeniable merit in the idea that computers can now be asked to perform actions that previously demanded manual cursor movement or typing. However, the same powerโusers who reap these benefits are also inadvertently overwhelming soft targetsโwebsites, interfaces, services, and people who have not yet decided how they wish to respond to the replacement of human customers, clients, and even friends with automated agents. The firstโmover advantage allows these agents to exploit weaknesses before defenses are erected, raising the question: once everyone possesses an โexceptionally capable personal agent,โ as Mark Zuckerberg has phrased it, will any untouched niches remain?
Government portals, brickโandโmortar stores, small contractors, and even dental insurers share a common trait: they lie far from the epicenter of the AI boom, have not solicited this transformation, and, for the most part, have not yet been compelled to consider how it might reshapeโor disruptโtheir daily operations. In contrast, organizations nearer to the AI core are experiencing change at a faster clip. Birju Kadakia, CEO of Rec, a startup that manages local sports and recreation for municipalities nationwide, told me that he is already observing early signs of a flood in SanโฏFrancisco. He noted a modest rise in agentic traffic within booking flows, emphasizing that this increase is not merely academic; large language models have been scanning the cityโs recreational pages en masse for years, and now they are attempting to complete transactions. Although the detectable share remains in the singleโdigit range even for activities popular with the most AIโsavvy populace, Kadakia suspects the true figure is higher because many agents operate indistinguishably from human browsers, eluding current detection methods.
Kadakia offered a concrete illustration: Muse helps him locate scarce pickleball courts and secures a reservation without requiring his constant supervision, and he is confident that numerous additional use cases remain unexplored. He went on to describe how booking platforms are evolving into deโfacto traffic cops, mediating between supply and demand in an environment increasingly populated by automated agents. Looking ahead, Rec envisions a world where perhaps half of all bookingโflow participants are some form of agent. To curb individual flooding, Kadakia proposes tying each agent to a single human identity, possibly through an agentic โpassport,โ although the industry has yet to converge on a standardized solution. For highโdemand slotsโsuch as the notorious 8โฏa.m. tennis openings in SanโฏFrancisco that have long been botโinfestedโhe suggests replacing the chaotic scramble with a randomized lottery drawn from a threeโminute window, thereby reducing the advantage of sheer volume. Furthermore, Rec could deploy its own agents to recommend underutilized facilities, guiding users away from overcrowded venues, and he imagines a future where agentโtoโagent collaboration enables partner matching and cooperative scheduling, turning competition into coordination.
Founders of Spot, an inviteโonly restaurantโreservation service that scrapes availability from platforms like Resy and OpenTable and then books tables, have echoed similar sentiments from a different angle. Andreas Larsen, one of Spotโs coโfounders, observed that NewโฏYork City has already been living in an AIโbot flood, and he believes that aggressive agents could ultimately serve a constructive purpose by pressuring booking platforms and restaurants to adapt and innovate. This pressure could open space for Spot to expand beyond its current niche of reservationโsniping into a fullโservice dining concierge that not only secures tables but also discovers and recommends eateries, thereby layering a new class of providers atop or between existing incumbents. Larsen recounted a personal anecdote in which Instinct managed his entire vacation itinerary, persuaded a distillery to open two hours early for a private tour and tasting, and booked the associated reservationโall without him sending a single email or placing a phone call. He concluded with a forwardโlooking vision: someday, agents might negotiate with one another to allocate global resourcesโsuch as coveted tables at Semmaโthrough a form of magnanimous superintelligence, though he remains uncertain whether such optimism will be realized.
Thus far, both outside and inside the tech sector, the arrival of AI has frequently been interpreted as a threat rather than an opportunity, prompting reactive measures that resemble floodโcontrol efforts instead of proactive redesign. In this mindset, todayโs Muse might successfully reach an insurance company, supply your information, wait on hold, and connect you to a human representative when the time is right, as documented by Eli Tan in the NewโฏYorkโฏTimes. Tomorrow, however, the same entity that invested heavily in constructing a frustrating phoneโattrition system could decide it no longer wishes to be the one shouting, โHuman, human, need to speak to human!โ A growing market already offers solutions to this dilemma: vendors advertise AIโcall detection tools that either permit useful automation through, demand verification, or shield human agent capacity; others promote โagent conciergesโ that intercept incoming AI traffic and allow it to proceed only within predefined limits. Importantly, many of the barriers that currently hinder AI are not accidental oversights but intentional design choices that can be reinstated or strengthened if stakeholders deem it necessary.
Navigating this emerging landscape requires a blend of vigilance, experimentation, and strategic foresight. For smallโbusiness owners, the first step is to establish clear protocols for handling unsolicited AI callsโsuch as training staff to politely verify caller identity, offering a callback option for legitimate requests, and logging interactions to detect patterns. Investing in basic callโscreening technology or partnering with services that provide AIโcall detection can reduce noise while preserving genuine customer engagement. Meanwhile, entrepreneurs and product teams should consider how to embed friction responsibly: implementing rateโlimits, CAPTCHAโstyle challenges, or identityโverification steps that deter abusive automation without frustrating real users. On the policy side, regulators ought to incentivize capacityโbuilding measuresโlike upgrading publicโservice infrastructureโrather than defaulting to restrictive barriers that stifle innovation. Finally, individuals can protect their personal data by routinely reviewing permission settings for AI assistants, limiting the scope of actions they can perform autonomously, and staying informed about the latest developer guidelines. By combining these practical steps with a willingness to adapt, stakeholders can transform the agentic flood from a disruptive surge into a manageable flow that enhances, rather than overwhelms, the systems we rely on.