Alex Holt, the chief technology officer at ElevenLabs, is about as bullish on AI agents as a C-suite leader can get.
The AI voice startup has deployed an inbound sales development representative (SDR) agent that has handled over 3,000 customer leads this year, qualifying over 90% of those potential clients in 32 different languages without any human intervention. Yet another AI agent serves as a coach to help new sales team members role-play customers interactions, resulting in faster onboarding for that part of the workforce. There’s also a recruiting agent that conducts first-round interviews with new job candidates, and within the company’s Slack, various agents are quick to jump in and answer any employee questions about internal processes.
“I think what’s most exciting about this space is that you know you always feel like you’re on the tip of the iceberg,” says Holt. He says that the efficiency gains ElevenLabs reaps from agentic AI may feel cutting edge today, but “in six months time, you can see even more gains.”
Increasingly, CTOs and chief information officers have homed in on agentic AI as a central focus for the value they expect to extract from their frothy investments in the technology. Agentic’s share of AI value grew from 17% in 2025 to 22% this year, according to research published by consulting giant BCG, and is projected to reach 39% by 2030. And yet, most surveys find that businesses are ill-prepared to rethink workflows and orchestrate multi-agent processes, in which one agent autonomously passes along work to another.
Holt says most employees at ElevenLabs work collaboratively with agents and a majority have even built their own agents to help handle repetitive tasks. The company’s core business objectives don’t change with AI: sales teams are expected to use AI to generate more revenue at a faster pace; for customer service, satisfaction scores are closely monitored.
“It’s exactly the same business metrics,” says Holt. “This has to be an initiative that the business units themselves are driving and pushing. The agents and tech are a tool that they can leverage to do that.”
Organizationally, ElevenLabs has deployed an engineer to sit on every team across the business, and every group has fewer than 10 people. Within recruiting, for example, two forward deployed engineers work in a pod with HR colleagues with the sole objective of reducing the time it takes to get candidates through the interview process.
Today, almost all prospective candidates are initially screened by an AI recruiter when applying for a role at ElevenLabs. “That’s 30 minutes that the recruiter can spend at the end of the process, convincing you to join,” says Holt.
Holt joined ElevenLabs in September 2023 as an engineer after spending five years at software and data analytics company Palantir Technologies. He was elevated to the CTO role in June 2026. Three months later, the startup announced that it doubled its valuation to $22 billion. CEO Mati Staniszewski recently indicated that ElevenLabs may begin to plan for an initial public offering in 2028.
The next frontier for agentic AI, says Holt, is to put more effort into multi-agent systems, enabling greater shared context and more complex work handoffs between multiple agents.
This is already under development for the company’s sales leads. The SDR agent uses voice AI to determine the seriousness of a prospective buyer, looping in a human representative after that screening process. But recently, one ElevenLabs customer was able to close a sale entirely through an interaction with two agents. First, the inbound request was fielded by the SDR agent, and later an account executive agent handled the legal documents and contracts.
“The value of that is you can now automate that whole chain for customers where it’s the simple, most repetitive mechanical cases,” says Holt. He says employees can dedicate more time to more complex inbound requests or upselling.
Holt says embracing agentic AI at this magnitude does require some patience. Is the recruiting agent helping to hire the right people? Are ElevenLabs’ customer service scores improving with AI? “Some of these metrics can take a long time,” says Holt. “We lean on the business owners to identify those leading indicators and the metrics we should be measuring.”
There are also instances where AI can act in unintended ways, an issue that’s raised alarm following recent high-profile instances in which OpenAI and Anthropic agents went rogue.
ElevenLabs says it has created a human-led monitoring and observation layer to govern its agents. But mistakes can still slip through. One customer service agent erroneously gave a small refund under $30 to a customer and failed to escalate the issue to an employee. ElevenLabs created a test case that simulated the call and the exact chain of thought and reasoning that led the agent astray, then updated the prompts and guardrails.
Holt says tracking every decision that an agent makes in this manner can actually be helpful. If a human made the same mistake, the company likely wouldn’t have noticed.
“It’s not just automating processes; it allows you to collect more data,” says Holt of agentic AI. “I expect over the next few years, the trend that we’re going to start seeing is people talking much more about optimizing these processes.”
John Kell
The post How ElevenLabs’ CTO is rolling out AI agents—and why he thinks bigger efficiency gains are ahead for the voice AI startup appeared first on Fortune.
