How ElevenLabs’ CTO is rolling out AI agents—and why he thinks bigger efficiency gains are ahead for the voice AI startup

ElevenLabs says its inbound AI sales agent handled more than 3,000 leads this year and qualified over 90% in 32 languages without human intervention; other agents assist with recruiting, onboarding, and employee questions. BCG research cited in the article puts agentic AI’s share of AI value at 22% this year, up from 17% in 2025, and projects 39% by 2030. ElevenLabs is developing multi-agent sales workflows, while acknowledging that agents can err: one customer-service agent issued an incorrect refund under $30, prompting updated tests and guardrails.

Analysis

The investable shift is from model capability to workflow ownership. As agents take on repeatable sales, recruiting and service tasks, value should accrue to vendors that already control customer records, permissions, escalation paths and audit trails—not necessarily to the model or voice layer. Salesforce, ServiceNow and NICE are plausible beneficiaries if they can embed agents in existing workflows; the same automation could pressure seat-based revenue and labor-intensive outsourcing as customers need fewer human interactions. Net impact depends on whether vendors capture usage or outcome-based fees before seat reductions dilute revenue.

ElevenLabs’ examples are useful proof points, not independently verified evidence of durable ROI: lead qualification and a completed low-complexity transaction do not establish conversion uplift, lower total cost, or customer satisfaction at scale. Treat its private valuation and potential IPO path as sentiment signals, not public-market catalysts.

Near term, agent announcements may support AI-software multiples, but the 1–3 month catalyst is measurable deployment economics: conversion, resolution quality, implementation cost and human escalation rates. Over 6–18 months, multi-agent handoffs widen the addressable task pool while multiplying failure points. Governance and evaluation tooling may become a requirement for procurement; a costly customer-facing error or regulatory scrutiny could delay deployments. The contrarian risk is that investors capitalize labor savings before enterprises redesign workflows or vendors demonstrate how those savings translate into recurring revenue.

Key Decisions for Investors

• No broad agentic-AI trade on this evidence alone. Track Salesforce, ServiceNow and NICE for disclosed agent adoption, usage-based revenue, seat trends and customer ROI; favor workflow/control-plane exposure only if monetization outpaces seat cannibalization.
• Watch for a relative-value opportunity: favor established workflow and contact-center platforms over undifferentiated agent offerings if deployments accelerate but buyers prioritize integration, permissions and auditability. Do not initiate until vendor-level adoption and retention evidence supports the thesis.
• For the next 1–3 months, monitor reported conversion or resolution outcomes, escalation/error rates, deployment costs and any customer-service or recruiting quality metrics. A rise in incidents, weaker customer satisfaction, or guidance that attributes seat declines to automation would falsify the bullish software thesis.
• Treat labor-intensive BPO and routine customer-support exposure as a 6–18 month downside watch, not an immediate short: delay in enterprise workflow redesign, compliance barriers or persistent human escalation could prevent expected savings from materializing.