At Campfire, we launched our consumer AI chat product, Cozy Friends, on Steam and mobile on October 12. The concept behind Cozy Friends was to combine the charm of games like Animal Crossing and The Sims with AI-driven interactions. In just the first 30 days, our users exchanged over 1.7 million messages with our AI agents, surpassing all of our expectations. This early success was not only gratifying but also confirmed that we had finally cracked the code for conversational AI products.
However, the path to success wasn’t easy. It took us an entire year of development, during which we had to build numerous tools, some of which we had to redo from scratch. There were many roadblocks along the way, and at times it felt like we were just banging our heads against walls. But those challenges taught us invaluable lessons and ultimately led to the creation of a product we were proud of.
Now, as we look ahead, I want to accelerate the process of building better AI chat products. I’m committed to helping others avoid the mistakes we made and the frustrations we endured. That’s why I’m open-sourcing all of our learnings and launching Sprites—our all-in-one tool designed to build, optimize, and scale conversational AI agents. Sprites is our attempt to streamline the process of creating effective AI chat products, and it’s built from the lessons we learned during our journey.
One of the most important lessons we learned is that AI chat outputs should be thought of as a complex function, not just a wrapper around a single large language model (LLM). This realization was key to improving our approach to conversational AI. In this post, I’ll share seven critical tips that helped us along the way, starting with the importance of system prompts. System prompts, like a dynamic application, need to evolve with the user’s journey and the state of the application, rather than being static. By treating your system prompt as a function of both user intent and application state, you can create far more engaging and effective conversations.