Zishen Wan Brings AI-Native Computing Research to Columbia

Zishen Wan, an assistant professor who joined Columbia University this fall, leads the Wan Lab researching AI-native computing, focusing on computing for AI and AI for computing. His doctoral work at Georgia Tech received the 2026 ACM SIGDA and FCCM Outstanding Ph.D. Dissertation Awards. He is teaching AI-Native Computing (COMS E6998) this semester.

Zishen Wan Brings AI-Native Computing Research to Columbia Computing systems now have to keep pace with increasingly capable artificial intelligence AI . Robots, agentic AI, autonomous systems, and other forms of physical AI demand more from the hardware and software that support them: systems that understand the structure, dynamics, and physical constraints of intelligent agents, and that can coordinate computation, memory, communication, and sensing across the entire stack. That challenge is at the center of Zishen Wan’s https://zishenwan.github.io/ research. Wan joined the department this fall as an assistant professor and leads the Wan Lab https://wan-research-group.github.io/ , where he studies two complementary directions: “computing for AI,” redesigning systems, architectures, accelerators, and chips for emerging workloads like embodied AI, autonomous agents, world models, and new forms of reasoning, and “AI for computing,” using AI itself to help design and optimize the computing systems and hardware underneath. “I think these two directions are becoming deeply connected,” Wan said. As AI systems grow more complex, designing the infrastructure beneath them is becoming increasingly difficult for humans alone, while AI is becoming capable of participating in the design process itself. That creates a feedback loop: better computing systems for AI, and AI that helps build better computing systems. “Studying both sides together gives us an opportunity to rethink not only what future computing systems look like, but also how they are created.” Wan’s current research builds on his doctoral work at Georgia Tech and postdoctoral work at Harvard. His dissertation, Tailored Computing: Cross-Layer System, Architecture, and Silicon Co-Design for Physical Intelligence, received both the 2026 ACM SIGDA Outstanding Ph.D. Dissertation Award and the 2026 FCCM Outstanding Ph.D. Dissertation Award. The work developed computing across the stack, from algorithms and software systems to architecture and silicon chips, with the goal of making emerging AI systems more efficient, reliable, and scalable. At Columbia, he’s extending it to the computing foundations needed for perception, reasoning, memory, and real-time control for the next generation of physical and agentic AI, as well as how AI agents can help design better computing systems. Wan sees opportunities to connect this work with colleagues across computer science, electrical engineering, AI, and robotics. Wan is bringing these ideas into the classroom this semester through AI-Native Computing COMS E6998 , a research-oriented course drawing students from machine learning, computer architecture, systems, and VLSI backgrounds. Rather than traditional assignments, students read recent research, discuss open problems, and develop semester-long projects on topics like running embodied AI models on edge devices, designing systems for AI agents, exploring accelerator architectures, and using AI agents to assist with computer architecture, RTL generation, optimization, and verification. “I want students to see these two transformations together,” Wan said. “The goal is to identify new systems and architecture questions created by AI, or new ways in which AI fundamentally changes how computing systems can be designed.” Looking ahead, Wan envisions AI-native computing systems in which intelligence is considered at every level, from applications and algorithms to architecture and silicon, with AI increasingly helping to design the systems that support it. Interested in AI-Native Computing COMS E6998 ? The course is aimed at graduate students and advanced undergraduates who want to explore how artificial intelligence is changing the way computing systems are built and used. Students can come from either an AI or computing background. The class emphasizes current research and open-ended problem solving, with students developing a semester-long project in areas such as embodied AI, agent systems, accelerator design, hardware optimization, RTL, or verification. It is a good fit for students interested in crossing traditional boundaries between machine learning, systems, architecture, and hardware.