AI Guru Jiang Yangqing Starts New Venture: Taking AI from Idea to Software Deployment

According to Dynasty Beating monitoring, it was revealed that about a month after leaving NVIDIA, LeptonAI founder Jia Yangqing, along with Bai Junjie and other former colleagues, founded Intent Lab. Their initial product, Fleet, is a self-managed software team composed of multiple Agents. Once users propose a goal, Fleet is responsible for design, coding, testing, validation, and post-launch continuous maintenance.

The most remarkable achievement is the GLM-5.2 inference engine. Fleet revamped TensorRT-LLM, increasing the output speed from 102 tokens/s to 647 tokens/s, a 6.3x improvement. Intent Lab claims that this is currently the fastest known GLM-5.2 inference speed.

Starting from a single prompt, Fleet also developed a database engine and passed 6 million SQLite compatibility tests. Another set of Agents added formal verification to the file system, automatically identifying and fixing data corruption issues. All three projects were completed by the same system.

Jia Yangqing believes that existing coding agents are already capable of quickly writing code; what is lacking is the ability to turn ambiguous requirements into complete designs and to provide upper-layer capabilities for long-term maintenance of production systems. Intent Lab is currently seeking its first set of enterprise clients, prioritizing projects that are excessively large in scale and have been idle for a long time.