2025 is often touted as the “year of the agent” especially given the convergence of several exciting trends: improving model capabilities (like reasoning models), hardware and engineering efficiency breakthroughs (like Deepseek), and billions in infrastructure investments across data centers globally (including Stargate’s US Package, Macron’s EU AI Package). 

More specifically, voice agents and voice models are emerging as the fastest growing new LLM ecosystem, with impressive launches from companies like Eleven Labs, OpenAI and Cartesia driving enterprise adoption. Text-to-speech models, speech-to-speech models, speech-to-text models have become more performant and accessible to end users. This is really exciting, especially for companies managing end-to-end customer interactions and applications that require lifelike voices that are accurate, reliable and responsive at scale.

In reality, AI agents struggle with reliability, especially when it comes to voice workflows at scale. Sure, you can connect to a speech model API, but you’ll have to tie those together with your own compound AI system– connecting tools like Pipecat or LiveKit to build your own orchestration engine that manages conversations and handles edge cases. And humans are discerning customers – the slightest robotic sound or latency issues can drive customers to hang up and abandon applications entirely.

Enter Phonic - the next generation speech-to-speech platform focused on reliability. We’re thrilled to lead their $4M seed round, joined by Lux portfolio company CEOs Qasar Younis of Applied Intuition, Clem Delangue of Hugging Face and Erik Bernhardsson of Modal. Inspired by their own experiences struggling to make voice agents work at scale, Phonic offers a one-stop shop platform to build voice agents - helping customers build, evaluate and observe their agents becoming the critical immune system to maintain complex agentic voice workflows. 

How do they do this? They’ve rethought the voice stack from the ground up - leveraging best-in-class realistic models they’ve trained from scratch, deploying an intelligent decision support system to protect against edge cases, and providing best-in-class latency (300 ms!). And in typical self-healing AI fashion, Phonic doesn’t just provide a great end-to-end experience, it continuously improves itself, becoming more intelligent with every interaction to make every end customer experience truly delightful.

Phonic is already making a real impact! Their platform is live in production with customers across industries like healthcare, customer support and e-commerce. Take Flexbone, one of their customers building AI agents for healthcare operations, where they’ve transformed their voice operations - simplifying complexity, ensuring hyperrealistic voice quality, and dramatically improving agent reliability. 

Most importantly, Phonic is spearheaded by an incredible team. We love backing technical founders with technical insights, identifying a core problem in the tech stack and innovating to create a unique solution. Phonic co-founders Moin and Nikhil epitomize that belief - they are a really special team that have been longtime friends of the Lux family. They first met as undergraduates at MIT and became fast friends and collaborators including co-founding the MIT Machine Learning Club together. Upon graduation, Moin joined MosaicML as one of their early research engineers helping close their first customer. Nikhil joined Genesis Therapeutics to spearhead their machine learning efforts. They quite literally grew up in the machine learning industry and are experts in training and running LLMs at scale. They’ve also recruited an incredible team - 5 out of their 7 team members are from MIT including several olympiad medalists!

As we move deeper into the “year of the agent,” Phonic is uniquely positioned to lead the voice agent revolution by solving the reliability challenges that have held back widespread adoption. Their comprehensive approach - combining proprietary realistic voice models, intelligent orchestration, and self-improving performance - creates the foundation for delivering on the promise of natural, effective voice interactions at scale. 

And Phonic is hiring for all roles! Check out their site here, sign up for a demo here and check out their jobs page to learn more.