ElevenLabs pushes AI agents deeper into enterprise work

By Olivia Hart

On November 11, 2026, ElevenLabs announced plans to add 15 megawatts of NVIDIA Vera Rubin compute capacity to its existing Blackwell Ultra systems. At its New York Summit, the company introduced model customization, the ElevenAI workplace app, a Partner Program and a Marketplace, extending beyond voice generation and transcription into customer-conversation management and employee work.

CEO Mati Staniszewski said conversational agents would become “critical infrastructure for every enterprise, as essential as cloud or payments” within a few years. He framed the products as a broader enterprise deployment push. ElevenLabs did not disclose figures on adoption, revenue, pricing or deployments in production, leaving the enterprise ambition largely unbenchmarked.

What does model customization change?

ElevenLabs said customization will begin in the coming months for selected open-source language models and its own voice models. Businesses will work with the company’s researchers to tune models for their customers and operating context, selecting metrics such as customer satisfaction or reliability. Under the plan, live deployment data lets companies keep refining models as interactions accumulate rather than relying only on a general industry model.

ElevenLabs already applies that industry-specific approach through Scribe v3 Medical, a healthcare transcription model built to recognize specialized terminology. Staniszewski said a domain-optimized financial-services model is also in development. The announcement did not provide performance benchmarks or a release date, and customization remains an unpriced proposal for buyers.

How does the Interaction Engine route work?

ElevenLabs’ previewed Interaction Engine manages turn-taking and interruptions. It can assign different language models to different parts of an exchange, using a cheaper model for triage and a frontier model for complex tasks.

ElevenLabs said the design avoids requiring a frontier model at every step and is intended to limit token use. The announcement did not identify the models, routing criteria or measured savings. The token-saving case remains unproven.

What do Memory and Routines add?

ElevenLabs said its ElevenAgents platform for deploying voice and chat agents will add Memory and Routines, with both features forthcoming. Memory will recognize returning authenticated customers and distill each conversation into a structured record of preferences, constraints and future occasions. The customer relationship management system will remain the primary customer record; in ElevenLabs’ example, a customer who prefers Spanish and can speak only after 5 p.m. would not repeat those details on the next call.

ElevenLabs described Routines as a way to automate multistep work before, during and after conversations by connecting existing tools and systems. Its examples include:

• pre-screening a mortgage applicant and calling back within minutes;
• reminding patients about visits and offering canceled slots to waitlisted patients;
• emailing reports, writing to a customer relationship management system and initiating calls.

Together, the features move ElevenAgents beyond answering inquiries toward retaining context and completing surrounding work. The company did not enumerate supported systems, error handling or governance requirements. It also provided no accuracy or retention terms for Memory, leaving any reduction in repeated questions unquantified.

How does ElevenAI serve employees?

Staniszewski previewed ElevenAI, an app that listens to spoken conversations, including in-person meetings, and produces notes, summaries and action items. The app can provide live assistance and pass context between people and agents. ElevenLabs said it takes meeting notes with participants’ consent and includes notice and consent mechanisms.

ElevenLabs opened free early access with the announcement. Dictation and meeting notes use Scribe v3, which the company describes as a recorded-audio transcription model covering 102 languages.

A separate Scribe v3 Live variant targets low-latency, real-time audio. The company did not set broader availability dates or explain how the app handles consent withdrawal, leaving the enterprise release undefined.

What does the partner network add?

ElevenLabs reports more than 550 partners across 44 countries, including Google Cloud, IBM, Deloitte, DXC and Genesys. The new Partner Program lets consultancies, systems integrators, software vendors and agencies integrate ElevenAgents into enterprise technology stacks.

Participants can also work with ElevenLabs’ forward-deployed engineers on complex deployments. An alliance with McKinsey and QuantumBlack, AI by McKinsey, pairs the platform with McKinsey’s enterprise-transformation work. ElevenLabs did not disclose partner revenue shares or minimum deployment commitments, so channel reach does not yet establish commercial scale.

The accompanying Marketplace initially offers a curated catalog of partner-built applications for healthcare, financial services, insurance, education and retail. Enterprises can browse the applications and contact their developers. ElevenLabs plans to start purchasing through existing commitments in 2027, so the channel remains a directory rather than a transaction system.

What does the NVIDIA expansion enable?

ElevenLabs and NVIDIA announced availability of Nemotron 3.5 Super VL in ElevenAgents. The company plans to add 15 MW of Vera Rubin capacity on top of its existing Blackwell Ultra systems, targeting large deployments with security or data-residency requirements. ElevenLabs also expects the capacity to give researchers access to optimized compute. No deployment schedule or utilization target accompanied the announcement.

What does Poland’s deployment show?

The expanded collaboration covers an on-premises voice agent for Centralna e-Rejestracja, the central appointment system in Poland’s national e-Health platform. EasyCall supports delivery of an agent that calls patients with only landlines to remind them about appointments and help them reschedule.

ElevenLabs said the central system handles 40 million appointments annually and has no-show rates of 10% to 20%. The deployment uses NVIDIA Confidential Computing to keep citizen data and audio within government infrastructure and protect it during processing. The project offers a concrete use case, though ElevenLabs supplied no call-volume or outcome data.

What must ElevenLabs prove next?

The announcement links domain tuning, persistent memory, workflow automation, employee assistance, partner distribution and specialized compute. Its execution test is narrower: move previewed features into broader production use, then convert Marketplace browsing into purchases through existing commitments.

The next proof points are broader production use and the planned start of Marketplace purchases in 2027. Without measured adoption, savings or reliability gains, ElevenLabs has not yet demonstrated that the connected package improves enterprise returns.