Voice has always been the missing piece in enterprise AI agent deployments. Text-based agents work, but they feel like software. Voice agents feel like people. On March 25, IBM and ElevenLabs announced a partnership that brings that gap one step closer to closing for large enterprises.

The two companies integrated ElevenLabs’ text-to-speech and speech-to-text technology directly into IBM watsonx Orchestrate, IBM’s enterprise agentic AI platform. The result is AI agents that can speak clearly across 70 languages, draw from a library of 10,000-plus voices, and handle the compliance requirements that enterprises actually care about.

What the Partnership Does

IBM watsonx Orchestrate is already used by companies in banking, insurance, healthcare, government, and utilities to build and deploy AI agents. Before this integration, those agents were largely text-first. The ElevenLabs integration changes the interface layer entirely.

Clients now get:

• Real-time speech recognition, so agents can understand spoken input and respond naturally
• 10,000-plus voices across 70 languages and regional accents
• PCI-compliant payment processing for voice-enabled transactions
• Zero Retention Mode for HIPAA-compatible data handling
• Data residency controls for organizations operating across different jurisdictions

Mati Staniszewski, co-founder of ElevenLabs, put the business case plainly: “AI agents are becoming central to everyday work, and voice is where AI either earns trust or loses it.”

Nick Holda, IBM’s VP of AI Technology Partnerships, echoed the shift: “We’re bringing a voice to AI agents in the enterprise.”

IBM’s approach here follows the pattern it used with NVIDIA on GPU infrastructure and Bain on post-quantum consulting. Rather than build voice capabilities in-house, it partners with the specialist. ElevenLabs is the specialist for voice quality and multilingual coverage. That’s a reasonable call.

Why This Matters Now

A few years ago, enterprise AI voice was dominated by rigid interactive voice response systems. Press one for billing, press two for support. Everyone hated them.

The first wave of AI-powered voice improved things somewhat, but the voices still sounded synthetic. Customers noticed. Employees noticed. Adoption was patchy.

What ElevenLabs brings to the table is premium voice quality that holds up in real conversations at scale. The difference between a robotic-sounding customer service agent and one that sounds human is not cosmetic. It is the difference between customers staying on the line and customers hanging up.

Combining that voice quality with IBM’s enterprise infrastructure, compliance layer, and existing client base creates something genuinely useful for large organizations that have been cautious about deploying voice AI because of how it would reflect on their brand.

What This Means for Business

For large enterprises already on IBM: This is a relatively low-friction upgrade. If you are already building agents in watsonx Orchestrate, you can now add voice capabilities without spinning up a separate vendor relationship. The compliance certifications mean less friction with your legal and security teams.

For mid-market businesses evaluating voice AI: The IBM-ElevenLabs announcement is a signal, not a buying decision. It tells you that voice AI is becoming standard infrastructure in enterprise stacks. If the biggest banks and healthcare systems are deploying voice agents, the technology has cleared the bar on reliability and compliance. The question for your business is not whether voice AI is ready. It is which deployment model fits your scale.

For businesses in regulated industries: The HIPAA and PCI compliance angle matters. Voice AI that can handle payment processing or health-related conversations without creating compliance exposure is a different product than a generic chatbot. If compliance concerns have held your organization back, this announcement narrows the list of legitimate objections.

For AI vendors and service providers: The partnership model here is instructive. IBM did not try to build world-class voice capability in-house. It found the best voice provider and integrated. That is increasingly how enterprise AI stacks get built. Best-of-breed components, orchestrated together.

The Bigger Trend

This announcement fits into a broader shift happening across enterprise AI. Twelve months ago, most AI agent deployments were text-only, used internally, and accessible to a handful of power users. The conversation was about what AI agents could do in theory.

Now the conversation is about interface quality, compliance, and scale. Enterprises are not asking whether to deploy agents. They are asking how to make those agents work well across channels, languages, and regulatory environments.

Voice is the channel where the user experience gap has been most visible. You can build a sophisticated AI agent that handles complex reasoning and integrates with fifteen different systems, but if it sounds like a robot on the phone, customers will not trust it.

IBM and ElevenLabs are betting that enterprise buyers are ready to close that gap. Based on what we’re seeing across the market, they’re right.

---

The practical next step is the free Working With Claude field guide. Thirty-two pages covering the ecosystem, Claude Code, and how to govern a rollout properly. Get your copy.