Joe Reeve

Joe Reeve is an engineer on ElevenLabs’ Growth team, helping developers build products with synthetic speech, transcription and conversational agents. His signature project, Talk to a Statue, transforms photographs of artworks into conversations with historical characters, exposing the difficult cultural and interaction-design questions beneath an engaging voice interface.

Earlier, Reeve served as chief technology officer of the Coronavirus Tech Handbook, where he built a real-time collaborative editing platform used by thousands of contributors. He founded Gived, an agency focused on socially beneficial research, design and development, and subsequently became an engineering manager at Amplitude. There, his interests included product analytics and making data governance more accessible with AI.

At ElevenLabs, he has written about integrating audio tools into Lovable, redacting sensitive information from speech transcripts and reducing text-to-speech latency. His focus is the practical gap between powerful audio models and products that developers can deploy responsibly.

• Talk to a Statue: Reeve’s multimodal cultural application photographs a statue, uses OpenAI to identify and research its subject, generates a character-specific voice with ElevenLabs’ Voice Design API, and opens a live ElevenAgents conversation over WebRTC. Existing APIs make the prototype relatively easy to scale; trustworthy museum experiences still require curator-shaped narratives and dependable collection data.

• Culturally responsible voice design: Assigning an artifact a voice involves judgments about its materials, origin, subsequent movement and present surroundings. Reeve treats accent and characterization as questions of historical interpretation requiring institutional expertise, not interchangeable product settings.

• Multimodal voice-agent interaction: Speech helps people express intent quickly, but lengthy spoken responses are cumbersome to navigate. Reeve favors interfaces combining conversational input with diagrams, generated interfaces or other information-dense visual output; he also imagines agents signaling when they want to interject.

• Interruption-aware conversational agents: If someone interrupts an agent mid-response, its stored history can wrongly include words the listener never heard. Reeve proposes using audio timestamps to truncate that history at the actual interruption, keeping subsequent responses aligned with the conversation as experienced.

Reeve also sees AI-assisted prototyping as an opening for people unfamiliar with conventional software design to invent unexpected interactions. His AI Engineer Europe session connects that experimentation with the craft of making technical ideas legible through concise storytelling, captions and carefully chosen music.