The Trillion-Dollar Industries AI Is Disrupting: Voice, Law & the End of the Billable Hour

Summary

This episode focuses on two fast-growing AI businesses attacking large, services-heavy industries: voice and legal. Max Junestrand says ElevenLabs started in 2022, released its first human-sounding text-to-speech model in early 2023, and then scaled from roughly $100M ARR to a reported $600M revenue/ARR context while keeping a research-heavy culture of small teams and no PMs. He emphasizes internal AI tooling, voice safeguards, licensed celebrity and estate deals, accessibility use cases, and a model-agnostic platform that sits above OpenAI, Anthropic, Google, and open source. The conversation then shifts to Legora, where the argument is that legal AI can reduce manual review, move junior lawyers from document grinding to agent orchestration, and pressure billable-hour economics. Junestrand’s core thesis is that winning in legal requires compliance, jurisdiction-specific data, and narrow models for narrow workflows rather than general-model ambition.

Notes

• ElevenLabs started in 2022, spent the first year on research/product, and released its first human-sounding text-to-speech model in early 2023.
• The company claims a steep revenue ramp: about 20 months to $100M ARR, then 10 months to $200M, five months to $300M, and now about $600M.
• ElevenLabs says it has 600 employees and is still optimizing onboarding and culture under fast growth.
• The company organizes around small five-to-ten-person teams and embeds engineers across functions, including legal, talent, and go-to-market.
• The first 10 employees reportedly had zero attrition, which the guest frames as evidence of strong research and engineering culture.
• ElevenLabs says it never had product managers, preferring people who can code, understand customers, and understand design.
• The company argues AI tools now let employees become much better across adjacent functions without waiting on specialists.
• ElevenLabs uses its own products internally, including an inbound AI SDR agent that can take calls instead of only web forms.
• The guest says voice UX is shifting toward quicker, more conversational turn-taking and more proactive assistants that know past interactions.
• He says the last 6-12 months brought a step change in voice-agent quality.
• WhisperFlow and Plaud-style note capture are presented as examples of voice input replacing tedious typing and manual follow-up.
• The guest likes Plaud Pocket because it captures conversational signal that would otherwise disappear and auto-fills notes and follow-ups.
• Voice input enables longer stream-of-consciousness prompting, which he says works unusually well with LLMs.
• In financial services, the company says AI voice reduces emotional friction for payment reminders and debt collection because people are more candid than with humans.
• ElevenLabs says it traces all generated content and moderates both voice and text to block scams and commercial misuse.
• The company says it has paid over $22 million back to talent through its voice marketplace.
• Voice licensing can be creator-priced or company-priced, depending on the use case.
• The company highlights accessibility as one of its most important applications, especially voice restoration for ALS and throat-cancer patients.
• ElevenLabs says its models preserve emotional tone across languages like Spanish, Italian, and Portuguese.
• The company says it worked with the James Earl Jones estate and Disney on a live interactive Darth Vader in Fortnite.
• Headspace is cited as a current public example of personalized meditation and interactive voice applications.
• ElevenLabs says it is model-agnostic and lets customers choose among Anthropic, OpenAI, open source, and Google models.
• The company’s strategy is to own the interaction layer and orchestration rather than any single foundation model.
• Max says ElevenLabs has outcompeted frontier labs on text-to-speech, speech-to-text, turn-taking, and music.
• He argues architecture and labeled audio data matter more than scale for voice model quality.
• ElevenLabs says it uses over 1,000 contractors to label audio assets.
• He expects voice to reach human-like Turing-test levels this year.
• The company says its growth has sustained about 50% quarter-over-quarter for seven quarters.
• Legal services are framed as a $1T market with only $40B in software spend, implying a 4% software share.
• He says law firms are experimenting with fixed-fee and success-fee pricing as AI changes incentives.
• Kirkland is cited as a ~$10B/year firm, illustrating how much value can sit in partner time.
• Legora’s legal engineer role is described as a forward-deployed lawyer who sits with major firms and helps them move from pre-AI to post-AI operations.
• The guest says AI is not just a mild productivity gain; it can do a lot of the work in legal practice.
• Junior-lawyer work is expected to shift from physical and virtual data-room review toward orchestrating agents.
• Legora says its data moat combines firm-owned precedent and organizational data with a global corpus of cases, laws, and regulatory updates.
• The product can adapt to local law, including cross-border matters such as a California GC serving a first customer in South Africa.
• Legora’s answer is designed to be immediately useful even if it is only about 80% accurate at first.
• He says legal research requires complete coverage, not just a top subset of cases.
• The old data-acquisition workflow is described as physically shipping books to India, scanning them, and preserving page citations.
• Agents after Opus 4.5 and 4.6 are described as able to perform intelligent case strategy and combine witness statements with case law.
• Legora says Claude’s legal offering is shallow and acts as a pipeline generator for the company.
• Max rejects general-model building and fine-tuning, favoring narrow models for narrow legal tasks.
• Legora’s tabular review workflow scales as documents times prompts, so 100 documents by 100 prompts becomes 10,000 API calls.
• Compliance is described as Legora’s core currency in selling to law.
• The company says it hosts sensitive contracts for national-security and weapons-manufacturing customers and works with governments.
• Legora says it does not do on-prem deployments because they slow roadmap execution.