Nicholas Leonard and Derek Caneja wanted to build AI voice agents, but they identified design flaws in the existing landscape. They observed that many agents were built with no-code tools, which allowed for fast production but often resulted in low-quality products. Conversely, other agents were created by companies with the resources to spend months building specialized tools. Leonard explained that developers and enterprises needed an alternative. He and Caneja also believed the future of software would be shaped by agents that code, validate, and optimize. These insights, along with a historical perspective, inspired them to create VoiceRun.
Leonard serves as the company’s CEO, with Caneja as the CTO. Last year, they launched VoiceRun, a platform that enables developers and coding assistants to launch and scale voice agents. While many low-code platforms rely on visual diagrams and prompt boxes to dictate agent behavior, Leonard noted this approach can be difficult to manage. VoiceRun takes a different path by letting users code their agent’s behavior directly, offering greater flexibility. Leonard explained that code is the native language for coding agents, allowing them to operate more effectively than within a visual interface.
Visual interfaces often have limited configuration options. For instance, creating a voice agent that speaks in a specific dialect can be challenging if the visual tool lacks that feature. In code, however, such a task becomes incredibly simple. Leonard pointed out there are countless small, custom tasks that visual interfaces simply cannot support.
Beyond enabling coding agents, VoiceRun also allows for A/B testing and one-click instant deployment. The company targets enterprise developers, helping businesses integrate AI into customer service or launch voice-based products. One example is a restaurant-tech company using the platform to create an AI phone concierge for handling food reservations.
VoiceRun recently announced the closing of a $5.5 million seed funding round led by Flybridge Capital. The AI agent space is highly competitive, with startups attracting significant investment. Leonard sees the company competing on two fronts: against no-code voice builders like Bland and ReTell AI for quick demos, and against sophisticated developer tools like LiveKit and Pipecat for maximum control. He positions VoiceRun in the middle, providing global voice infrastructure and an evaluation-driven lifecycle while letting customers retain ownership of their business logic code and data.
Leonard emphasized that the key difference is closing the loop for end-to-end coding agent development. The expectation is that developers will supervise coding agents that write code, run tests, deploy, and propose improvements.
Ultimately, Leonard hopes his product will help create voice agents that make people more comfortable with automated systems. Currently, customers often feel relief when a human answers the phone, as existing voice automation can be brittle and ineffective. A survey from last year found that three-fourths of consumers still prefer human customer service. Leonard aims to change this perception, noting that human agents have their own limitations, such as language barriers or making customers feel judged.
He drew an analogy to the automotive industry, stating that while there were great cars before the Model T, vehicles did not become ubiquitous until the assembly line was invented. Similarly, while great voice agents exist today, they will not become ubiquitous until the voice agent factory is built. Leonard believes VoiceRun is that factory.