Calling customer service is becoming a strange experience because the voice answering may sound completely human without belonging to one. ElevenLabs CEO Mati Staniszewski believes businesses should tell customers when AI voice agents are handling those conversations, at least for now.
The position is notable because ElevenLabs provides the technology behind a growing number of automated customer interactions. Its voice systems support companies including Klarna, Deutsche Telekom, Cisco and Adobe, while governments are also deploying its technology for public services.
During an interview in Toronto, Staniszewski said customers currently need disclosure because most people still expect another person when they make a phone call. Without that warning, discovering that a highly realistic voice belongs to software can leave customers feeling misled.
However, he does not believe that expectation will last forever. Staniszewski predicts that within roughly five years, personal AI agents could become common enough that people will routinely expect machines on both sides of phone conversations.
That possibility takes the discussion beyond whether AI voices sound realistic enough to pass for humans. It raises a bigger question about what happens when businesses and customers increasingly send software agents to speak for them.
The timing is particularly relevant as AI companies move beyond chat interfaces and give agents more independence. Google is already testing Gemini calls that can contact businesses, navigate phone menus, make reservations and reschedule appointments for users.
Meta is moving in the same direction with Muse, its personal agent designed to operate across email, shopping, computers and other services. As these systems spread, AI voice agents may increasingly find themselves speaking directly with other AI agents.
AI Voice Agents Are Becoming Harder to Distinguish From People
ElevenLabs has built much of its business around making synthetic speech sound less mechanical and more emotionally convincing. Staniszewski believes the next major challenge involves understanding emotions during a conversation rather than simply reproducing realistic speech.
He said the company’s longer-term ambition is to pass what he described as the conversational AI version of the Turing test. That would require an agent to recognize emotion and adjust how it speaks, perhaps slowing down when someone sounds confused or changing its delivery when frustration rises.
ElevenLabs already markets agents capable of handling customer service, scheduling and other business processes across phone, web and digital channels. Its platform supports conversational agents in more than 70 languages and integrates with systems including Salesforce, Zendesk, Stripe and Twilio.
This is where disclosure becomes important today. A robotic voice immediately signals that someone is dealing with automation, while increasingly natural AI voice agents can remove that obvious distinction.
Staniszewski argues that businesses should currently make the difference clear rather than allowing customers to discover it midway through a conversation. He suggested that companies could offer people a choice when human support carries a substantial waiting time.
For example, a customer facing a 30-minute wait could choose between remaining on hold or speaking immediately with an AI agent. Staniszewski said people presented with that option usually select the automated service and are often surprised by its quality.
That approach makes transparency part of the customer experience instead of presenting AI as a hidden replacement for employees. It also allows customers to decide when they are comfortable handing a conversation to software.
The situation becomes more sensitive in industries where a mistake can affect money, health or other important decisions. Staniszewski said simpler informational calls can often rely on open-weight models, while financial services still require stronger models because mistakes carry greater consequences.
ElevenLabs is already working in areas where those differences matter. Its enterprise products connect voice agents to internal tools and APIs, allowing them to resolve issues, schedule appointments and complete other actions rather than simply answering questions.
The company is also working with Customers Bank on AI agents for account inquiries, card services, transfers and customer onboarding. ElevenLabs says those systems include routes to human employees when conversations require additional judgment.
The Bigger Shift May Be Agents Talking to Agents
Staniszewski’s most interesting prediction is that disclosure could become less important because people themselves may stop making many routine calls. Instead, their personal AI agents could handle those conversations automatically.
In that world, telling someone that the customer service representative is an AI agent becomes almost redundant. The caller may already be an AI agent representing a human customer on the other side.
Google’s latest Call for Me experiment offers an early glimpse of that possibility. A user can ask Gemini to contact a business, wait through automated menus and carry on a conversation using approved personal information.
Meta similarly wants Muse handling shopping, appointments and other tasks across outside services. Both developments suggest consumer agents are becoming capable of communicating with companies instead of simply preparing information for their users.
That creates an unusual future for customer service. A personal agent could eventually call a company’s support agent, explain a problem, authenticate an account, negotiate an available solution and return only when human approval becomes necessary.
Human callers might rarely hear the realistic voices that companies are spending considerable money developing. Voice technology could instead become part of an invisible machine-to-machine negotiation happening in the background.
For ElevenLabs, that does not necessarily reduce the opportunity. Businesses would still need reliable conversational systems capable of understanding requests, accessing information and completing transactions whether the other participant is human or artificial.
The company says it has now reached $600 million in annual recurring revenue, with more than 55% coming from traditional enterprise customers. Much of the remaining business comes from smaller companies, developers, builders and creators.
Its official figures showed ARR above $500 million by May, after ending 2025 at $350 million. ElevenLabs attributed much of that acceleration to businesses deploying voice agents across support, sales, recruitment and marketing operations.
Growing adoption also means companies must decide where transparency fits as automated conversations become more realistic. Being technically capable of sounding human does not automatically mean customers want uncertainty about who or what is speaking with them.
Staniszewski’s position acknowledges that tension while suggesting social expectations could change quickly. People who find AI customer service uncomfortable today may behave differently once their own agents regularly perform similar tasks on their behalf.
That transition could also change what customers value during support calls. Waiting for a human might become less attractive if an AI voice agent can solve an ordinary problem immediately and transfer complicated cases when necessary.
ElevenLabs is betting heavily on that shift, even if doing so puts pressure on its margins. Staniszewski said the company is willing to pass savings to customers and accept lower margins when that helps demonstrate the technology’s value and expand adoption.
For now, though, his position is straightforward: people should know when the voice on the other end belongs to artificial intelligence. The more interesting question is how long that disclosure will still feel necessary once both sides of the conversation have their own agents.
