Over the course of the summer, I noticed an incremental yet frustrating rise in my monthly internet bill—approximately fifteen dollars more than what I had been accustomed to paying. Rationally, I knew there were avenues available to counteract this increase: a quick comparison of competing service providers could almost certainly yield a more competitive rate. And yet, despite being aware of this relatively simple solution, I repeatedly postponed making the necessary phone call. My hesitation stemmed not from complacency but from sheer exhaustion—born of all-too-familiar experiences involving labyrinthine automated menus, interminable hold times punctuated by repetitive music, and synthetic voices that failed to comprehend my increasingly desperate pleas to “speak with a human representative.” The thought of navigating that ordeal again felt disproportionate to the potential savings it promised.
Nevertheless, a novel idea emerged: what if I could offload this tedious responsibility to someone—or rather, to something—else, without paying a cent? My curiosity led me to experiment with a generative AI voice tool. Through it, I created a digital persona—an “agentic” version of myself—tasked specifically with negotiating my internet bill down to a fairer price. I endowed this virtual proxy with limited patience and a directive to be assertive, a design choice that inadvertently resulted in a brusque, occasionally condescending tone. Listening back to its recorded exchanges made me wince in sympathetic embarrassment for the genuine customer service representative on the other end. My AI agent frequently repeated or paraphrased the rep’s own words with exasperation before expressing disappointment, then inventing imaginary lower rates purportedly offered by competitors to use as leverage. At one point, it proclaimed indignantly, “I’ve been a loyal customer for years, and it’s unjust that newcomers get better deals,” following up with the biting challenge, “Are you suggesting I should just cancel altogether?”
In today’s era of rapid AI proliferation, this role reversal is more than ironic. As customers struggle to connect with actual human employees, human employees now face the challenge of distinguishing genuine clients from sophisticated AI impersonators. Call center operators increasingly find themselves conversing with artificial voices crafted through advanced deepfake technologies—simulations so convincing that the line separating person from program grows faint. The consumerization of such AI tools presents a dual-edged possibility: while they can be misused for fraudulent or malicious intent—such as mass trolling campaigns or data manipulation—they can also serve as an everyday convenience for overworked, digitally weary individuals seeking relief from the bureaucracy of customer service interactions.
“We’re merely witnessing the early stages of what could become a substantial and far-reaching issue,” warns Brian Levin, chief customer officer at Reality Defender, a company that develops software capable of detecting AI-generated content in real time for governments, financial institutions, and other large organizations. Reality Defender, as part of my research, guided me through the process of constructing my synthetic voice agent. Levin noted that innovation in the realm of automated, agentic callers is outpacing the ability of traditional contact centers to adapt or regulate the influx of such technology.
Echoing this analysis, Emily Fontaine, the global head of venture capital at IBM—both an investor in and partner of Reality Defender—remarked that AI voice agents are increasingly targeting help lines across industries. Their proliferation, she explained, is fueling not only a rise in fraudulent activity but also an overwhelming surge in total call volume, further taxing infrastructures already burdened by high demand. Likewise, Patrick Carroll, founder and CEO of the deepfake detection firm ValidSoft, observed that contact centers are now fielding a growing number of calls from artificially generated voices designed to circumvent security checks and authentication systems. Even corporations fortified with robust defense mechanisms, he said, are experiencing dramatic escalations in automated call attempts as the accessibility and sophistication of voice generation software expand.
In many ways, customer service departments now find themselves ensnared in an AI paradox. Research conducted by Gartner, a leading IT analytics firm, predicts that by 2029, artificial intelligence will handle as much as eighty percent of routine customer requests, potentially slashing operational costs by nearly one-third. Yet despite this promise of efficiency, a separate Gartner survey revealed a mismatch between expectations and reality: while eighty-five percent of customer service executives report either implementing or exploring AI chatbots, a mere eleven percent claim these systems have delivered results approaching full satisfaction. This technological balancing act—between innovation and instability—creates fertile ground for exploitation by fraudsters and opportunists, sparking what some experts have called an impending “AI vs. AI” arms race in the customer service sector.
As these tools become more user-friendly, the temptation to create personalized “negotiator bots” to contact one’s cable or utility provider will likely become irresistible to consumers. Daniel O’Sullivan, a senior director analyst at Gartner, explained that many customers actively desire this kind of outsourcing to avoid the stress of negotiating personally. Businesses, accordingly, will have to adapt to the reality of dealing with both human and artificial callers. However, O’Sullivan cautioned that integrating AI-driven customer representatives into such systems carries notable risks. Companies would be required to enhance their ability to identify fraudulent or malicious calls, brace for heightened contact volumes arising from reduced communication barriers, and confront the potential erosion of human rapport—those fleeting yet valuable connections fostered by empathetic agents who go beyond procedural assistance.
The desire to bypass long call queues is hardly new. According to another Gartner survey conducted in 2025, nearly half of consumers now attempt to resolve service issues indirectly—searching on platforms like Google, Reddit, or even ChatGPT—before resorting to contacting a company directly. The impulse to outsource this frustration has existed for over a decade; as far back as 2010, a company called LucyPhone offered a simple yet ingenious way to avoid time on hold by automatically reconnecting users once a live representative became available. Later, Google’s “Reserve with Google” tool extended the principle to restaurant bookings, while modern platforms such as DoNotPay have elevated it further. The latter employs AI to fight fines, contest tickets, and recover owed funds, serving as a surrogate advocate against bureaucratic systems designed to be unintentionally—or strategically—obstructive.
Creating my own AI agent, however, proved neither instant nor effortless. It demanded nearly an hour of precise setup: cloning my voice using audio samples from a past radio interview, feeding it my account credentials, and ensuring it understood those details as thoroughly as if they were second nature. Although this trial did not save time, the refinement of voice-generation technology was striking, particularly when compared to my earlier attempt at a similar experiment with my bank six months prior. Back then, I had to manually generate every possible utterance in advance, anticipate likely questions, store all responses as separate audio files, and trigger each playback manually during the interaction. The newer tool, by contrast, allowed me to define personality traits, constraints, and goals, and then release the system to converse dynamically in real time—almost like directing an unscripted performance.
Matt Smallman, a prominent call center security specialist, noted that building such systems is now within reach for curious hobbyists or technologists who simply wish to make a statement about automation. “People are understandably anxious about what lies ahead,” he remarked, emphasizing how easy it is to foresee a world in which these tools become so ubiquitous that both pranksters and practical users rely on them to handle every mundane, clerical, or transactional interaction—from disputing healthcare charges to scheduling maintenance for household appliances.
Ultimately, my digital counterpart failed to convince the company representative to restore my rate to its original promotional price. The employee explained politely but firmly that their employer did not match competitor promotions and that no alternative discounts were available after introductory offers expired. Still, the exchange was striking. Over the five-minute conversation—tense yet civil—my AI agent insisted repeatedly on fair treatment, feigned outrage, and even threatened account cancellation to pressure the representative into concessions. The employee, seemingly unaware of the caller’s nonhuman nature, responded as though addressing an irritated but genuine customer.
When it finally requested escalation to a supervisor, I intervened quickly, worried that my digital proxy might take its mission too literally and terminate my essential internet service. The experiment underscored a fundamental limitation of such automation: while it can mimic determination and articulation, it lacks contextual awareness and emotional judgment. For now, I am not quite ready to delegate my advocacy entirely to an algorithmic counterpart. I’d rather pick up the phone myself, speak courteously, and appeal to the empathy of another human being.
Amanda Hoover, senior correspondent at *Business Insider*, reports on major technological trends and the shifting dynamics of the digital economy. Her column in *Discourse* reflects on pivotal technological transformations shaping society, offering analysis grounded in experience, reporting, and critical insight.
Sourse: https://www.businessinsider.com/scammed-internet-provider-lower-my-bill-deepfakes-ai-2025-11
