Building Viral AI Experiences with Vibe Coding

The 'Talk to a Statue' Pipeline

Joe Reeve, a growth engineer at ElevenLabs, built a viral application that allows users to photograph a statue and engage in a real-time voice conversation with it. The technical architecture relies on a rapid, multi-step pipeline: a photo is taken, OpenAI's Deep Research identifies the subject and generates historical context, ElevenLabs' Voice Design API creates a custom voice profile based on that context, and an ElevenLabs agent manages the conversational flow. The entire process executes in approximately 30 seconds.

The Power of 'Vibe Coding'

Reeve emphasizes that the project was completed in two hours on a Sunday using Cursor, an AI-powered code editor. He defines 'vibe coding' as the ability to describe desired outcomes to an LLM rather than manually writing boilerplate code. This approach allows developers to focus on the 'glue'—the narrative and interaction design—rather than solving low-level technical challenges. The success of the app, which garnered 1.5 million impressions, highlights that for consumer AI, the story and the interaction pattern are often more important than the underlying technical complexity.

Designing for Voice and Interaction

Reeve discusses the philosophical and technical challenges of voice interfaces. He notes that current voice agents often feel 'tacked on' rather than integrated into the environment. He suggests that the future of voice interaction lies in multimodal experiences where users can see and manipulate the agent's 'thought process' or context alongside the audio. He also highlights the 'politeness problem,' where users are hesitant to interrupt AI agents, suggesting that better interaction design is needed to encourage more natural, interrupt-driven conversations.

Scaling and Future Directions

While the prototype was built quickly, scaling it for museums requires moving beyond generic web research to curated, high-quality data. Reeve notes that museums possess significant IP in their databases, and the real work lies in collaborating with curators to define the 'voice' and 'narrative' of inanimate objects. He suggests that the next evolution of consumer AI will likely mirror the social-graph-based viral growth seen in early Facebook games, where simple, shareable primitives drive mass adoption.