(Illustration: Taken at The Venetian, where I get up early every year-end to enjoy a sunrise breakfast. Chef Thomas Keller built two restaurant brands: The French Laundry, the embodiment of perfection, and Bouchon, the same obsession over detail but made approachable for everyday life. ElevenLabs has a similar structure. Research with real power shouldn’t be locked away in papers or experimental data; it belongs in products, where users walk past it every day. The deepest craft eventually walks toward the everyday. Image source: Ernest.)
✳️ Tech Can Be Chased Alone, But the Org Structure With Its Culture Is Hard to Copy
Research-driven companies often hit the same problem: the technology works, but the product won’t move. Or the reverse, the market need is crystal clear, but the research can’t keep up with the rhythm. ElevenLabs started in 2021 as a weekend experiment by two Poles, Mati Staniszewski and Piotr Dabkowski, focused on AI voice synthesis, with applications spanning audiobooks, dubbing, voice agents, and games, where the synthesized voices can almost fool a human. In just a few years, they’ve reached an 11 billion USD valuation with a team of over 300 people. They aren’t just better at the technology. The way their organization runs, fused with their culture, becomes the moat. Tech alone can be chased, but the org structure together with the culture cannot easily be copied. From a16z’s interview with CEO Mati Staniszewski, three lines of thought are worth observing in detail. (Of course, every family has its own difficult scripture; lines of thought have to match constraints and context. Don’t copy them straight into your own situation after reading.)
1️⃣ Let research and product spin together, not hand off in waterfall style
- In many companies, the research team finishes a round of work and packages it up for the product team to implement.
- There’s a time lag in between, a translation cost, and direction drift. By the time research output reaches product and then customers, the market may have already moved.
- ElevenLabs’ approach is to let the product side feed user signals directly to the research side, while the research team deploys new models directly into production for testing.
- Real-time bidirectional feedback. It isn’t a relay handoff, it’s a flywheel. (A note from someone who’s been there: as soon as there’s an interface between two sides, you still need to watch for dropped balls; gentle reminders and tracking mechanisms between teams are still necessary.)
- Mati said some companies have research but no product, and some have product but no research. They try to have both, and let both sides accelerate each other. When the product is built by the people from research who actually care, users can feel that investment.
(This reminded me of what Boris from Claude Code recently said about needing more “generalists” in the future. Ten years ago, when I was helping draft Bluetooth international specifications, I had a similar realization. Once a spec document is written and released, it needs continuous feedback from implementers to refine. Some usage scenarios and edge cases, when first written into the draft, hadn’t been implemented by any vendor yet. We had to wait for the first wave of implementer reports to know whether the wording was rigorous enough to be understood. And many of the first-hand implementers were chip vendor engineers, not necessarily the industry application engineers familiar with the spec. It’s precisely because of that bidirectional feedback loop between spec and implementation that vendors around the world could build mutually compatible products. ElevenLabs’ research-product flywheel is structurally the same thing. That’s also why, when we were planning our own org back then, we merged product and research into a single department. A decade and change ago, very few people understood it, but maybe we accidentally picked up a winning hand (?). I’m grateful to everyone who’s given us advice along the way to refine it.)
2️⃣ Use “no titles” as a filter, not just a culture statement
- ElevenLabs removed all traditional titles, including VP-level ones.
- Mati said this isn’t to look cool; it’s a deliberately designed filter mechanism:
- People who want titles hear about it and don’t apply, naturally filtering out the high-ego candidates.
- Those who stay don’t avoid asking questions, sharing ideas, or giving advice just because someone else outranks them.
- Engineers can directly access training clusters and run experiments on their own ideas, no layered sign-offs needed.
- The team doubles every six months, but because the small teams have high autonomy, it doesn’t feel like the company is getting bigger.
(Speaking of which, when I was at TSMC I was lucky to be young and bold enough to ask questions, and before graduation I’m grateful that my manager sent me to a cross-fab service team under a certain VP to practice communication and coordination. All those layers of accumulation became the foundation for the ports and portals I’m building today.)
3️⃣ Hire for proof of excellence, not traditional resumes
- ElevenLabs hired heavily from non-traditional backgrounds in the early days.
- The team includes people with astrophysics backgrounds, applied physics master’s degrees, someone who worked at the White House for the President, and even someone ranked in the top 250 of the European Dota leaderboard.
- Mati said they look for “some proof of excellence”, which can be an open source project, or something done outside of work.
- It doesn’t have to be a fancy degree, but you need to show that you’ve gone deep into something and produced results.
- They apply a rigorous cultural-fit screen to make sure rapid scaling doesn’t dilute team values.
(Every consulting firm’s report says enterprises plan to increase AI investment in the future, but Accenture’s research shows that fewer than one in five enterprises actually has modernized workflows. Most organizations just treat AI as a tool, without thinking about how to change the workflow itself. ElevenLabs is exactly the counterexample. They redesigned from the org structure level, exploring the nature of the workflow, which matters more than rolling out tools.)
For existing organizations, don’t get stuck on having to remove titles. Look at whether there’s a real-time feedback channel between your R&D and product teams. If every piece of feedback has to go through meetings and sign-offs to reach the other side, the flywheel won’t spin. Try shrinking one small feedback loop to under a day, for example by having two departments share a folder or a meeting, and feel the difference. Org design doesn’t need an overnight overhaul, but feedback channels and access permissions can be opened up first. The speed difference starts there.
📷 Caption 👉 Taken at The Venetian, where I get up early every year-end to enjoy a sunrise breakfast. Chef Thomas Keller built two restaurant brands: The French Laundry, the embodiment of perfection, and Bouchon, the same obsession over detail but made approachable for everyday life. ElevenLabs has a similar structure. Research with real power shouldn’t be locked away in papers or experimental data; it belongs in products, where users walk past it every day. The deepest craft eventually walks toward the everyday. As Frieren’s Himmel was asked, “What does it take to be remembered?” he answered, “Even the smallest thing, try to change someone else's life. That alone is enough.”
The first turn of the flywheel is always the hardest, just as hard and as small as taking a moment every day to leave a “like 👍” or a “heart ❤️”. Try a few more taps, and once it spins, the everyday view becomes a sky-high overview of the whole landscape.
✳️ Further Reading
- The Man Who Built Claude Code Was Changed by It
- Why Enterprise AI Deployment Gets Stuck: Lessons from Mistral AI’s Approach
- Coexisting with AI: Three Restructuring Principles from Block
- Before Claude Managed Agents: Breaking Down Palantir’s Five-Layer Framework
- 2025 Year in Review: Slow Down to Go Fast
✳️ Knowledge Graph
(More about Knowledge Graph…)
✳️ Transcripts
The Evolution and Future of Voice Interfaces
- We’ve been trying to make these human voices for literally since the 1700s.
- Then in the early 1900s, we had the first digital synthesizers.
- » Are you um » that it doesn’t cross that threshold of actually sounding like a human and actually making you feel something.
- Then it shifted into Siri, which has kind of a bit of back and forth.
- that sounds more realistic, but again, it doesn’t cross that threshold of actually sounding like a human and actually making you feel something.
- » Mati, it’s so great to have you here at the headquarters of Andreessen Horowitz.
- » No, thanks so much for having me.
- It’s incredible to to be here and speak here together about some of the work we do.
- You have said voice is uh poised to become the next fundamental interface for humans interacting with computers just as mouth touch screens and keyboards.
- Help us imagine what it looks like.
- » A lot of things are screen first.
- Most people will have the laptop, the phone most of the day in front of them.
- I think a lot of that will move into the background where where you will be able to be a lot more present.
- When I imagine say studying in a classroom in the future, you have on headphones you can have the most smart physicist, mathematician, a historian helping you through learning the the subject.
- There will be an interesting shift where voice will be a big part of the technology where today when you go to other countries, other cultures, you you cannot fully immerse inside the culture unless you know the language.
- And with voice and with technology suddenly this will will become possible where you can speak any language in the world and fully understand the not not only what is said but how it’s said kind of feel a closer part which which will be just incredible future where the the true language barriers but also the cultural barriers or the things that we have never learned will become possible.
Origins and Early Growth of ElevenLabs
- » Let’s start at the beginning.
- You and Piotr grew up in Poland.
- Tell us the experience that sparked idea of ElevenLabs » in Poland.
- If you if you watch a foreign movie, all the voices where it’s a male or female voice are narrated with one single character.
- So, one voice speaks all the lines.
- » They ought to make the day that » what?
- » Well, it’s 8:00 and it’s not a good day.
- » All the emotionality, all the intonation just just disappears.
- And then back in 2021, we realized that it’s still happening.
- Piotr was at Google, I was at Palantir.
- We would explore different projects together on the weekends and we invited the first group of users then and started kind of iterating a little bit deeper and then we started getting good signal on what are some of the use cases that that that will really resonate.
- So when we launched in early January we already have a a few thousand people lined up that we knew are very interested in actually using the product but then of course the few thousand turned into a few hundred thousand of users and that was a magnitude probably higher than like we expected in the first order.
ElevenLabs Product Philosophy and Strategy
- Introducing voice design V3.
- Introducing ElevenLabs image and video.
- » Proudly introduces studio 3.0.
- » What has been the guiding principle of the product philosophy?
- » It always was a combination of a where do we think we can deliver value with some of the research uh uh work but then layering the product on top.
- Two, where do we think there’s actually real problem?
- Like there are companies who have the research, there are companies who have the product and we try to have the both and I think we have it's great because product can directly talk to the kind of provide the feedback what is needed to the research research then immediately is able to to iterate on that they can also test their models directly on the product and with this way you know it just like the both kind of accelerates.
Rapid Team Expansion
- » Talking about the team you went from just the two of you around the pre time to I believe seven people when you’re raising the series A and did the launch and then quickly to a few dozen a year later.
- How did you approach imbuing the team?
- What qualities are you looking for when you’re hiring?
- » We were like especially in the early days hiring from very non-traditional backgrounds.
- » So I did astrophysics in my undergrad and then applied physics in my masters.
- » Yeah.
- So I first met Mati when we did a hackathon together when we were 21.
- » I was working at the White House for President Biden and an ElevenLabs investor told me that I should do everything in my power to try to go work there.
- I was always pretty ambitious, but like most of my ambition I put into video games.
- I have like 12,000 hours of Dota or something.
Non-Traditional Backgrounds
- I was actually like ranked 250 or something on the on the European leaderboard.
- days trying to hire for some proof of excellence that that people would do and and it could be an open source project.
- It could be doing something outside of work.
- » Yeah, I was doing my master’s degree.
- I wasn't really going to university much.
- I was developing this text of speech project and like kind of like Piotr wore me through a guitar.
- » When I finished my thesis, I posted online one of the samples of the music generation model and Piotr saw this this example and contacted me.
- So when I first joined, we had a 11 desk room.
Global Hiring Philosophy
- Now we have offices in over 11 cities, over 300 employees.
- We’re doubling every 6 months.
- But because we're remote first and we work in very small teams with high autonomy, you actually forget how big the company is.
- We wanted to hire the best people in the in the world and we don't think there's that many researchers in the world that are at that top level especially in in voice maybe 50 maybe 100.
- So we we wanted to hire wherever wherever they are.
- » There is as you know this like very strong cultural obsession to be in person.
- How do you contrast these two different setups?
Global Vision and Culture
- » When we started the aspiration was very global both in in what we wanted to create as as a technology.
- We wanted to make it available across all languages, across all geographies.
- » ElevenLabs had a culture when I came in and I think that was also what enticed me.
- I understood the vision that Mati and Piotr had for what type of company they want to build and the type of people that they are, which essentially is reflected in the culture.
Founder Personalities and Strengths
- » Mati and Piotr, they’re childhood best friends.
- They know each other super well.
- They’re both incredible operators and they’re high trust.
- Honestly, what really got us excited about investing in the company was was chatting with the founders, Mati and and Piotr.
- They had a really unique vision of like what the world could look like in the future that a lot of people didn't see yet.
- » Mati and Piotr are like yin and yang in a way.
- Piotr is very focused on the research.
- He’s an absolute genius in that space.
- » Working with him is is is very nice because he’s is very technically can go very technically in depth.
- The second smartest person I know is significantly less smart than him.
- Like let’s let’s put it like that.
Team Dynamics Analogy
- It is a bit like good cop bad cop.
- Maybe maybe kind of like Mati is the good cop and Piotr as the bad cop.
Evolving Leadership Role
- Well, we’ll start a little bit with uh thanking you for being here.
- You’re hard guy to catch.
- So uh where are you today?
- » Now I’m in Dubai.
- » How has your role evolved as you became larger more remote team?
Autonomy and Flat Hierarchy
- Yeah, it’s definitely you don’t know all engineers which is definitely sad that you at some point you just will not know all these people in the company.
- Mati knew everybody at the previous offsite already on on the previous offsite when there were 100 people already failed.
- If you have great people, there’s very little effort needed to to run the company because you can just trust these many founders.
- People that that really take ownership and care about the company because you because you love working here and you love the product.
- When the product is built out of love, then users can see that » everyone’s very high autonomy.
- They’re low bureaucracy, very flat, fuzzy hierarchy.
- They’re doing whatever is needed to move the needle for the customers to ship quickly.
- We removed all the titles, and it's a great way, both initially, of filtering for people who are very low ego.
- And so if you’re coming in, “Yes, I want to be VP of blah, blah,” you’re not going to get VP.
- And so it actually will turn off those people.
- But I’d argue that’s a good thing.
- No implicit bias of asking a question, or asking for help, or giving advice to someone, or proposing ideas because there’s no explicit hierarchy.
- Get access to a training cluster and train a model that you have an idea for.
Ensuring Cultural Fit in Hiring
- » We are applying a rigorous screen for ensuring a cultural fit before we bring someone in.
- And I think that's essential to being able to scale this quickly and still preserve culture.
- In fact, when I first spoke about this publicly and we, we kind of launched the idea that we got rid of titles, I had someone that I used to work with reach out and she said, like, “I heard that you got rid of titles.
- I love that notion.
- What roles do you have?
- I want to join.”
- And she’s now leading hiring, incredibly successful.
Specialized Audio Models
- » Currently, we have specialized models for audio, for sound effects, and for music.
- And I think the future of sound is kind of like having one model which can generate any kind of audio.
- You could imagine seeing something with voice that is converted to music, or like singing something, changing the, the singing into sound effects.
The Vocal Turing Test and Raw Audio
- The new challenge we've really set ourselves is, can we be the first company to cross this threshold of the vocal Turing test?
- How do you have an AI which really sounds like a human that you can interact back and forth with, but is super smart, super empathetic?
- I think there's going to be a point where most of the communicate we do with machine might be through audio because one, it's faster to communicate, but also because it's more information-rich.
- There's things now that machines or LLMs are not capturing.
- If you train a model on text, you’re basically using text units, tokens that are, uh, created by humans.
- Where if you train a general audio generation model, you’re training on, on raw audio.
- If you can make a model that is smart in audio, you can imagine you can make a model that is smart in any raw data domain.
- That, I think, is one of the most interesting things.
Emotional Impact of Voice
- Voice is the only AI modality that can actually make you feel something.
- And so when you have text, yes, you can have a poem or a story, but it doesn’t give you that same kind of emotive feel.
- Well, as when you hear a voice, whether it’s like a ASMR whispering voice, or whether it’s a deep booming cinematic voice, it can really kind of transport you and make you feel, make you feel alive.
Personal Motivation and Defining Future Technology
- » I love to end the conversation with this question.
- What drives you personally?
- Definitely seeing people react is, is one of the, always the best moments.
- But I, I feel like I’m in just such a lucky position where, you know, I can, I can work on a company with my best friends.
- But now it feels like we have this, this incredible team of somewhere between sports team to family where just everybody is driving on the same passion and and and and vision.
- But I think now especially, it’s, it’s just so rare that you get a chance to be the voice of the change, or voice of the technology, and be able to be at the frontier and, and define how voice will be that interface for everybody around us.
- It’s just such an unique opportunity to, to create a bike that, uh, that we are lucky and happy to, to be able to, to be part of it.