Conviction

Trajectory builds a platform for continual learning, letting companies post-train open-source AI models on real user interactions so products improve weekly instead of staying static between major model releases. Founded in 2025 by former Google DeepMind, Apple, OpenAI and Meta Superintelligence Labs researchers, it works with AI-native customers including Decagon, Clay and Harvey to ship narrowly-tuned models that outperform frontier labs on customer-specific tasks.