Character.AI

Before ChatGPT, before Gemini, before the world knew what a large language model was — there was Character.AI. Noam Shazeer and Daniel De Freitas left Google in late 2021 not because they were bored, but because they were furious. Google had let them build Meena, then LaMDA — some of the most capable conversational AI systems in the world at the time — and then refused to release them to the public, citing safety concerns that Shazeer and De Freitas believed were being used to slow-walk technology that could genuinely help people. So they left. Despite personal pleas from Sundar Pichai to stay, they walked out of one of the most generous compensation environments in tech to work in a rented office in Palo Alto, building from scratch a platform that would let anyone talk to any character they could imagine.

The product they built was not just technically impressive — it was emotionally resonant in a way that no AI product before it had achieved. By December 2022, three months after public beta launch, users were generating one billion words per day on the platform. Twitter and Reddit were full of people describing how Character.AI had helped them work through social anxiety, battle loneliness, practice languages, and process grief — talking to AI versions of deceased family members, therapist personas, historical figures, fictional companions. Users were spending an average of 75 minutes per day on the app — more time than Instagram, more time than TikTok for many demographics, more than any other AI application on earth. Then, in August 2024, Google brought the founders back in a reverse acqui-hire arrangement valued at approximately $2.5 billion in licensing and talent — one of the most significant AI talent acquisitions in history. Shazeer joined Google DeepMind to co-lead Gemini. De Freitas joined as a research scientist. About 30 employees followed. Character.AI continued independently under Karandeep Anand, hired as CEO in June 2025 with a mandate to commercialize and stabilize the platform after the founders' departure. The safety chapter has been difficult: lawsuits from parents of minors who were harmed following interactions with platform characters, FTC investigations, and the November 2025 decision to remove open-ended chat for users under 18 — replacing it with curated non-chat experiences including Feed, Imagine, AvatarFX, and Streams. The platform that pioneered AI companionship is now navigating the most consequential question in consumer AI: what does a responsible, sustainable AI social platform actually look like?