I remember sitting in a cramped meeting room back in 2016, pitching MovieLaLa to a room full of skeptical investors. We were building something cool, but the tech felt clunky. Fast forward to today, and the conversation has completely shifted. I've backed over 200 startups, including heavyweights like OpenAI, Anthropic, Scale AI, and Hugging Face. Let me tell you something. Voice AI is the next frontier. It's not just a neat trick anymore. It's fundamentally changing how we interact with machines.
When we sold RemoteTeam to Gusto, one of the biggest challenges we faced was communication across distributed teams. Typing is slow. Reading is slow. Talking is fast. That's why I'm so bullish on conversational AI. We are moving away from screens and keyboards. We are moving toward a world where you just talk to your computer, and it understands you perfectly.
Let's talk about speech recognition. For years, it was terrible. You'd ask Siri to set a timer, and she'd call your ex. It was a joke. But the models have gotten insanely good. The error rates have plummeted. I see this firsthand in the pitch decks crossing my desk every week. Founders are building applications that can transcribe meetings with near-perfect accuracy, even with heavy accents or background noise. This isn't just convenient. It's a massive productivity boost.
The Rise of AI Voice Cloning
Here is where things get wild. AI voice cloning. A few years ago, generating a realistic human voice took hours of audio data and massive compute power. Now? You can clone a voice with a three-second clip. I've seen demos that blew my mind. The inflection, the emotion, the pacing. It's all there.
This opens up a massive can of worms, obviously. Deepfakes are a real problem. But the upside is enormous. Imagine personalized audiobooks read by your favorite author. Imagine customer service bots that actually sound empathetic. Imagine being able to license your voice for commercials or video games without ever stepping into a studio. The monetization opportunities are staggering.
I always tell founders in my "Becoming Top 1%" community to look for the non-obvious applications. Everyone is building a ChatGPT wrapper. Don't do that. Build something that solves a real, painful problem.
For example, think about accessibility. Voice AI assistants are giving a voice to people who have lost theirs due to illness or injury. That is powerful. It's not just about making money. It's about making an impact. I recently spoke with a founder who is using voice cloning to give ALS patients their original voices back. That brought tears to my eyes. That is the kind of technology I want to fund.
Conversational AI is the New UI
We are entering an era where the primary user interface is conversational. You won't click through menus or type queries into a search bar. You will just ask for what you want.
I saw this shift coming when I invested in Hugging Face. The open-source community is driving innovation at a breakneck pace. The big tech companies are trying to lock everything down, but the open ecosystem is moving faster.
If you are building a product today, you need to think about voice from day one. It can't be an afterthought. It needs to be baked into the core experience.
Here are a few things I look for when evaluating voice AI startups:
- Latency: If there is a noticeable delay between when I speak and when the AI responds, the illusion is broken. It needs to be instantaneous. I want sub-500 millisecond response times. Anything slower feels like talking to a robot on Mars.
- Contextual Understanding: The AI needs to remember what we talked about five minutes ago. It needs to understand nuance and sarcasm. If I say "Yeah, right," it needs to know I'm being sarcastic, not agreeable.
- Emotional Intelligence: Can the AI detect if I'm frustrated or happy? Can it adjust its tone accordingly?
These are hard problems to solve. But the founders who figure them out are going to build massive companies.
The Reality of Building in Voice AI
Let me be blunt. Building a voice AI company is brutal. The compute costs are astronomical. The talent is scarce. The competition is fierce.
When I was building RemoteTeam, we had to be incredibly scrappy. We didn't have unlimited resources. We had to focus on the things that actually moved the needle. The same applies here.
Don't try to build a foundational model from scratch unless you have a billion dollars sitting in the bank. Use the existing APIs. Fine-tune them for your specific use case. Focus on the user experience. That is where you will win.
I see too many founders getting bogged down in the technical details and forgetting about the customer. Nobody cares about your parameter count. They care about whether your product solves their problem. I had a pitch last week where the founder spent 20 minutes talking about their transformer architecture and zero minutes talking about who was going to buy the damn thing. I passed.
The Evolution of Speech Recognition
Let's dig deeper into speech recognition. When I first started angel investing, speech-to-text was a novelty. It was something you used when you couldn't type, but you fully expected to go back and edit the garbled mess it produced. I remember trying to dictate an email to an investor while driving down the 101. The result was so bad I almost crashed my car laughing.
Today, the environment is entirely different. We have models that can understand context, distinguish between multiple speakers in a crowded room, and even pick up on industry-specific jargon. This is a massive leap forward.
Think about the implications for customer support. For decades, call centers have relied on rigid, frustrating IVR systems. "Press 1 for sales. Press 2 for support." It's a terrible user experience. Now, we can deploy conversational AI agents that actually understand what the customer is saying. They can handle complex queries, process returns, and even upsell products, all without human intervention.
This isn't just a cost-saving measure. It's a revenue generator. When you provide a better customer experience, people buy more. It's that simple.
Voice AI Assistants in the Enterprise
The consumer applications of voice AI are obvious. Alexa, Siri, Google Assistant. We all know them. But the real money is in the enterprise.
When we were scaling RemoteTeam, we had employees in 15 different time zones. Coordinating meetings was a nightmare. We relied heavily on asynchronous communication. But text has its limits. It lacks tone. It lacks nuance.
Imagine an AI voice assistant that sits in on all your meetings, takes perfect notes, identifies action items, and automatically updates your project management software. That's not science fiction. That's happening right now.
I've invested in several companies that are building exactly this. They are taking the friction out of collaboration. They are allowing teams to focus on deep work instead of administrative tasks.
And it goes beyond just meetings. Imagine a voice interface for your CRM. Instead of clicking through dozens of screens to update a lead, you just say, "Update John Doe's status to closed-won and send him the onboarding sequence." Boom. Done.
This level of automation is going to unlock massive productivity gains across every industry. I'm seeing startups that are saving sales teams 10 hours a week just on data entry. That is real ROI.
The Ethics of AI Voice Cloning
We need to talk about the dark side of this technology. AI voice cloning is incredibly powerful, but it's also incredibly dangerous.
I've seen the deepfakes. I've heard the synthetic audio clips of politicians saying things they never said. It's terrifying. As an investor, I have a responsibility to think about the ethical implications of the technologies I fund.
We need robust authentication mechanisms. We need ways to verify that a piece of audio is authentic. We need clear regulations around the use of synthetic voices.
But we can't let fear stifle innovation. The potential benefits of AI voice cloning are too great to ignore.
Think about the entertainment industry. Actors can license their voices for video games, animated films, and audiobooks. They can literally be in two places at once. This opens up entirely new revenue streams.
Think about education. We can create personalized learning experiences with AI tutors that speak in a voice the student responds to. We can translate educational content into hundreds of languages instantly, with perfect pronunciation and inflection.
The key is to build these technologies responsibly. We need to prioritize transparency and consent. If we do that, the upside is limitless.
My Investment Thesis for Voice AI
People always ask me how I pick winners. How did I know to invest in OpenAI or Anthropic before they were household names?
The truth is, I don't have a crystal ball. But I do have a framework. When I look at a voice AI startup, I'm looking for three things.
First, I look at the team. Are they obsessed with the problem? Do they have the technical chops to execute? Are they resilient? Building a startup is a grind. You need founders who won't quit when things get hard. I want founders who have been punched in the face by the market and kept going.
Second, I look at the data. In the world of AI, data is oxygen. Does the company have access to a unique, proprietary dataset? If they are just relying on publicly available data, they have no moat. They will get crushed by the incumbents.
Third, I look at the distribution strategy. Having a great product is not enough. You need a way to get it into the hands of users. Do they have a clear path to market? Do they understand their customer acquisition costs?
If a company checks all three boxes, I write a check. It's that simple.
The Future of Human-Computer Interaction
We are at an inflection point. The way we interact with computers is fundamentally changing.
For the past forty years, we have been forced to speak the computer's language. We had to learn how to type. We had to learn how to use a mouse. We had to learn how to navigate complex graphical user interfaces.
Now, the computer is learning to speak our language.
This is a profound shift. It's going to democratize access to technology. It's going to empower people who have traditionally been left behind.
My mom still struggles to use her smartphone. She gets confused by the menus and the icons. But she can talk to Alexa just fine. She can ask for the weather, play her favorite music, and set reminders.
That is the power of voice AI. It removes the friction. It makes technology invisible.
Lessons from the Trenches
I've learned a lot of hard lessons over the years. Building MovieLaLa and RemoteTeam taught me that the best technology doesn't always win. The best user experience wins.
When we were building MovieLaLa, we spent months perfecting our recommendation algorithm. We thought it was the best in the world. But our users didn't care. They just wanted a fast, easy way to find movie trailers. We had over-engineered the product.
I see the same thing happening in the voice AI space today. Founders are obsessed with building the most advanced models. They are chasing state-of-the-art performance on obscure benchmarks.
But the user doesn't care about benchmarks. The user cares about whether the product works.
If your voice assistant takes five seconds to respond, the user is going to abandon it. If your speech recognition software constantly misinterprets common words, the user is going to abandon it.
You have to obsess over the details. You have to polish the user experience until it shines.
The Role of Open Source
I mentioned earlier that I invested in Hugging Face. I did that because I believe deeply in the power of open source.
The big tech companies want to control the AI ecosystem. They want to lock developers into their proprietary platforms.
But open source is the great equalizer. It allows anyone, anywhere in the world, to build on top of state-of-the-art models. It accelerates innovation. It drives down costs.
We are seeing an explosion of open-source voice AI models. Whisper from OpenAI was a massive catalyst. It proved that open-source models could compete with, and even beat, proprietary solutions.
This is creating a massive opportunity for startups. You don't need to spend millions of dollars training a foundational model. You can take an open-source model, fine-tune it on your specific dataset, and build a world-class product.
Navigating the Hype Cycle
Let's be real. There is a lot of hype in the AI space right now. Every company is slapping "AI" on their pitch deck and hoping for a massive valuation.
As an investor, I have to sift through the noise. I have to separate the signal from the hype.
When a founder pitches me a voice AI product, I ask them one simple question: "What happens when the novelty wears off?"
Voice cloning is cool. Talking to a virtual avatar is cool. But novelty doesn't build a sustainable business. Utility builds a sustainable business.
Your product needs to solve a real problem. It needs to save people time. It needs to save people money. It needs to make their lives better in a tangible way.
If you can't articulate the utility of your product, you don't have a business. You have a science project.
The Hardware Challenge
We can't talk about voice AI without talking about hardware.
The software is advancing rapidly, but the hardware is lagging behind. We need better microphones. We need better edge computing capabilities. We need devices that can process voice commands locally, without relying on the cloud.
This is critical for privacy and latency. If every voice command has to be sent to a server in Virginia to be processed, we are never going to achieve the instantaneous, seamless experience we want.
I'm keeping a close eye on startups that are building specialized silicon for voice AI. I think there is a massive opportunity there.
The Global Opportunity
Voice AI is not just a Silicon Valley phenomenon. It's a global opportunity.
In many parts of the world, literacy rates are low. Text-based interfaces are a barrier to entry. Voice AI can bridge that gap. It can provide access to information, education, and financial services to billions of people who have been excluded from the digital economy.
I'm incredibly excited about the potential for voice AI in emerging markets. We are going to see entirely new use cases and business models emerge from these regions.
The Next Five Years
I've been in the tech industry for a long time. I've seen trends come and go. I've seen hype cycles inflate and burst.
But I have never been more excited about a technology than I am about voice AI.
It's not just another feature. It's a fundamental shift in how we interact with the digital world. It's going to disrupt massive industries. It's going to create entirely new categories of products.
If you are a founder, this is the time to build. The tools are available. The market is ready. The opportunity is massive.
Don't wait for the perfect moment. Start building today. Be scrappy. Talk to your users. Iterate rapidly.
And if you build something amazing, you know where to find me.
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.