By Teresa Torres · producttalk.org · @ttorres on X · LinkedIn
Teresa Torres interviews the founders and CTO of Snapbar, a 14-year-old company that started as a wedding photo booth business, pivoted to a virtual product when COVID eliminated its events, and then rebuilt itself around generative AI. The conversation traces how the team moved from WebRTC-based capture to Stable Diffusion, custom LoRA fine-tunes, and reasoning-model prompting for brand-safe generated images and video. It also covers their experiential marketing platform that lets brands create branded 'worlds' at live events, and their agent orchestration approach to engineering with coding agents. The episode is a case study in how a necessity-driven pivot can become an applied-AI advantage built on domain expertise rather than model access alone.
01Key takeaways
- Necessity can force a rebuild from first principles that later becomes a strategic advantage.
- Domain expertise applied to AI often creates more value than access to the underlying models.
- Custom fine-tuning and careful prompt design can produce outputs competitors cannot easily match.
- Give users participatory roles in generated experiences to turn a product into a co-creation moment.
- Structure coding agents around context, tools, verification, and workflows to ship faster with a small team.
- Build brand-safety and representation checks into the generation pipeline rather than treating them as afterthoughts.
02Key sections
- Origins and the pandemic pivot
- The company began as a side hustle in wedding photo booths and grew into a national events business. When COVID canceled events overnight, the team rebuilt the experience as a cross-platform virtual product from first principles.
- Adopting generative AI
- Generative AI entered the product through Stable Diffusion, with the team fine-tuning custom models to reach brand-quality output. Their prompting evolved from negative prompts to longer reasoning-model prompts that give brands finer creative and safety control.
- Representation and brand safety
- A meta-prompting preprocessing step aims to represent users' likenesses accurately, including non-obvious details. Brand-safety controls shape what gets generated so outputs stay on-brand.
- Experiential platform and participation
- Brands use the platform to 'world build' at conferences and events, inviting fans into branded creative experiences. Mad Lib-style inputs and prompt injection turn photo moments into co-creation between brands and audiences.
- Building with agents
- The small team uses Claude Code and Codex to ship quickly and has developed an agent orchestration framework built on context, tools, verification, and workflows. They are now letting brand customers configure experiences themselves through agent-driven vibe coding.
03From the post
“Listen to this episode on: Spotify | Apple Podcasts What does it take to reinvent a 14-year-old company—not once, but twice? In this episode of Just Now Possible, Teresa Torres talks with Sam Eitzen (Co-founder & CEO), Joe Eitzen (Co-founder & CPO), and Patrick Ellis (CTO) of Snapbar”
04Frameworks mentioned
Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.