David SenraBuilding One of AI’s Fastest-Growing Companies | Mati Staniszewski, ElevenLabs
CHAPTERS
- 0:02 – 2:56
Launching ElevenLabs before ChatGPT: an AI-native company from day one
David and Mati open by contrasting “this time is different” entrepreneurship narratives with what it actually meant to start ElevenLabs in 2022. Mati explains their backgrounds (Palantir/BlackRock and Google research) and the founding goal: unify frontier research with real-world deployment.
- •Started ElevenLabs in 2022 before ChatGPT; benefited from focus while hype was on crypto/metaverse
- •Mati’s product/customer background at Palantir; Piotr’s deep research background at Google
- •Company thesis: combine frontier research with product deployment under one roof
- •Early ambition: build a platform that transforms how organizations communicate
- 2:56 – 5:58
Why audio first: Poland’s “single narrator” dubbing problem and the Babel Fish vision
Mati traces the initial inspiration to Poland’s low-quality “one voice narrates all characters” localization style. That frustration became a bigger vision: breaking language barriers and enabling natural voice-based interaction with technology.
- •Polish media localization used a single narrator voice, stripping emotion and quality
- •Initial product concept: high-quality dubbing with original emotion/intonation preserved
- •Long-term idea: real-time cross-language understanding (Hitchhiker’s Guide “Babel Fish”)
- •Early uncertainty: knew modalities would expand, but not how fast interactivity would arrive
- 5:58 – 12:37
From dubbing to text-to-speech: creators’ needs reshaped the roadmap
To deliver dubbing, ElevenLabs would need strong transcription, translation, and speech generation—none were good enough in 2022. Creator feedback also revealed more immediate pain points (editing lines, pre-reading scripts, narration), pushing the team to focus first on best-in-class speech generation.
- •Dubbing pipeline requires transcription → translation → regenerated speech
- •State of the art was robotic/unstable; components weren’t production-ready
- •Creators wanted practical tools first: narration, script readthrough, line fixes
- •Strategic pivot: solve speech generation quality first, then expand to full dubbing
- 12:37 – 16:15
“Research + product deployment” as the company identity (and why audio-only on research)
Mati defines ElevenLabs as a hybrid: a frontier audio research lab plus a deployment platform for communication. He argues that research focus—specifically audio—creates architectural advantages and defensibility, while product integrates audio with knowledge and LLMs.
- •Company framing: frontier audio research (TTS/STT/orchestration) + one communication platform
- •Use cases span marketing, support/call centers, sales qualification, and operations
- •Research focus on audio: needs both science and art; quality is subjective and hard to benchmark
- •Belief that architectural breakthroughs (not just data/compute) are key in audio
- 16:15 – 18:47
Focus as a moat: saying “no” to video-first products and choosing the right intersections
Mati explains how they decide what not to build: if audio isn’t the bottleneck, they avoid the opportunity. He gives an example of pausing avatar/lip-dubbing efforts when video quality was the limiting factor, revisiting later when ecosystem models improved.
- •Decision filter: do we have unique advantage via audio? If not, don’t pursue
- •Example: avoided building avatar/lip-dubbing when video was the core bottleneck
- •Revisiting multimedia now as open-source video components mature
- •Preference: integrate audio into existing assets rather than build end-to-end text-to-video
- 18:47 – 21:23
Building an ecosystem around voice: marketplace, authentication, and revenue sharing
Beyond model quality, ElevenLabs is building ecosystem advantages—trust, brand, and a marketplace where creators can monetize verified voices. The marketplace expands voice variety while aligning incentives through compensation per usage.
- •Ecosystem strategy complements research defensibility over the long term
- •Voice marketplace: users create voices, ElevenLabs authenticates, others can license/use
- •Revenue share: voice owners get paid when their voice is used (e.g., “George” in Eleven Reader)
- •Scale: tens of thousands of marketplace voices with diverse accents/styles/languages
- 21:23 – 24:09
From models to platform to applications: making “communication” the unifying theme
David presses on product sprawl; Mati clarifies the unifier is communication across the customer journey. ElevenLabs aims to be one of a few core interaction platforms, while explicitly avoiding competing in broader “intelligence/knowledge/coding” domains.
- •AI pattern: model → platform → applications; ElevenLabs embraces it with a single theme
- •Unifier: on-brand communication across marketing, sales, support, and operations
- •Positioning: not building general intelligence/coding platforms; focus on interaction layer
- •Goal: become a leading platform where companies configure knowledge, integrations, and brand once
- 24:09 – 26:54
Deutsche Telekom case study: marketing → call center agents → in-network translation
Mati walks through a deeply integrated customer example showing multi-product expansion over time. Deutsche Telekom began with marketing content, then added call-center voice agents, and later deployed in-network agent experiences including real-time translation during calls.
- •Initial entry point: marketing content/podcasts in DT’s app using ElevenLabs voices
- •Expansion to support: call-center voice agents integrated with DT knowledge systems
- •Further deployment: agent inside the network to join calls, schedule, and translate
- •Closed-loop learning: capturing interaction data to improve customer journey end-to-end
- 26:54 – 33:39
Forward Deployed Engineers (FDEs): the Palantir playbook for integration and learning loops
Mati explains why FDEs became critical: enterprise voice agents require real integrations, testing, monitoring, and iteration. He credits Palantir’s culture—small teams embedded with customers—and emphasizes that FDEs should feed lessons back into the core product, not become pure services.
- •Enterprise deployment requires CRM, telephony (SIP/Twilio), knowledge, flow design, testing
- •FDEs work on-site with partners; treat customers as collaborators
- •Key distinction: FDEs are part of product, not just go-to-market
- •Second-order value: reuse learnings across customers to accelerate product R&D
- 33:39 – 37:19
Flat orgs, transparency, and AI-native management: wider spans and better information flow
Drawing again from Palantir, Mati describes a low-title, flat organization with high transparency. He expects AI to improve information flow so leaders can access ground truth directly, manage wider spans of control, and shift from reactive to proactive operations.
- •No-title/flat org philosophy; minimal layers of management
- •Extreme transparency: broad access to docs and data across the company
- •AI as an internal “information router” summarizing granular activity for leadership
- •Wider spans (~10 direct reports) enabled by better summarization and visibility
- 37:19 – 42:49
Small teams + engineers everywhere: embedding technical leverage in every function
Mati predicts a broader trend: non-technical teams will increasingly include engineers to automate workflows and elevate AI adoption. He notes centralized enabling functions still exist (legal, sales enablement, RevOps), but are scaled with engineering and internal tooling.
- •Small, autonomous teams adopt AI bottom-up without heavy mandates
- •Engineers embedded in ops, talent, and legal to automate and raise capability
- •Centralized enablement still matters: legal, sales enablement, revenue engineering
- •Dogfooding: internal agents and tooling to scale processes without headcount bloat
- 42:49 – 44:30
Where growth is strongest: agents, creative, and fast-moving industries (fintech leads)
Mati breaks down where revenue is coming from and which sectors are adopting fastest. Conversational agents and creative tooling are core growth lines, with fintech moving quickest, followed by healthcare, telcos, and retail/e-commerce.
- •Primary revenue drivers: agents and creative product lines
- •Fintech adoption is fastest (e.g., Revolut, Klarna)
- •Next wave sectors: healthcare and telcos; retail/e-commerce accelerating this year
- •Regulatory complexity varies by sector but demand is broad
- 44:30 – 55:19
Taste, art, and the “human” in AI voice: authenticity, imperfections, and identity
The conversation shifts to why voice is uniquely powerful and technically hard: it encodes emotion, identity, and nuance. Mati notes even voice agents perform better when they include human-like imperfections (ums, pauses), reinforcing that “taste” and craft matter alongside engineering.
- •Voice carries emotion/intonation/imperfections that create connection and authenticity
- •Agents improved when they became less “perfect” and more human-like
- •Audio modeling is harder to benchmark due to subjective voice preference differences
- •Building in AI requires both art and science; “taste” becomes a differentiator
- 55:19 – 1:04:01
Entrepreneurship path, partnership with Piotr, and the decision not to sell
Mati reflects on becoming a founder as a non-obvious path in Poland and how weekend projects with Piotr evolved into a company. He discusses acquisition interest, explains why they’re committed to staying independent, and frames this era as a rare chance to build something lasting.
- •Entrepreneurship felt less culturally common in Poland; confidence grew through career exposure
- •Mati and Piotr collaborated via hack projects for years before committing
- •Received several acquisition offers; turned them down and plan to stay independent
- •Motivation: this is a unique historical window; biggest risk is not trying
- 1:04:01 – 1:11:07
Voice as the interface for AI—and turning conferences into a business tool (not the circus)
Mati argues voice will be a major interface for accessing intelligence, enabling more natural, emotionally aware interactions. He closes by explaining why he attends select conferences: when prepared, they function as concentrated partner/customer meeting hubs and expansion engines.
- •Near-term future: AI interactions become more conversational, emotionally aware, and voice-led
- •Analogy: like the telephone vs telegraph—no need to “learn the machine’s language”
- •Babel Fish as a North Star: enable translation/communication in existing devices
- •Conference strategy: do few, prepare heavily, pre-arrange meetings, use for sales/expansion
