September 8, 2026 · MIT Technology Review
AI researcher Danijar Hafner builds startup on agents that plan ahead
Danijar Hafner, known for his work on world models and reinforcement learning including the Dreamer series, has quietly founded a new startup in San Francisco. The stealth-mode company is developing AI agents designed to plan ahead for unexpected situations rather than just react.
Why it matters: Hafner's background in model-based reinforcement learning and world models puts him at the center of a push to make AI agents more robust in unpredictable, real-world settings, a persistent weakness of today's LLM-based agents that mostly react step by step. A well-known researcher launching a company around this problem suggests planning and world-modeling techniques may be moving from academic benchmarks toward commercial agent products.
Related updates
- Kyutai releases CADRE video autoencoders for low-latency world modelsOct 23
- Study: fancier coding-agent harnesses cost more, don't score betterOct 24
- Google privacy policy now allows ad targeting from connected Gmail and Photos dataOct 23
- Anthropic's Claude generates machine-checked correctness proofs for 123 open-source librariesOct 23