In this video
What you'll learn
Build a working agent live, no slides
Watch a real AI agent get built on Kimi K3, from empty file to working.
See Kimi K3 run a real workflow end to end
Watch the new 2.8T model handle a genuine build, not a canned demo.
Copy the exact stack for your own builds
Leave with the setup and steps to rebuild the agent on your own projects.
Why this topic matters
Moonshot shipped Kimi K3 on July 16: 2.8 trillion parameters, a 1M-token context window, and a debut at #1 on LMArena's Frontend Code arena. At $3 per million input tokens, the agent workflows that were too expensive to run all day become affordable. In this session I build a working agent on K3 end to end, live and no slides, then hand you the exact setup to rebuild it yourself.
You'll learn from
Nathan Wang
OPC Founder · 18 AI products in production · PhD Engineer/TPM Apple & Meta
Building the operating stack for the One-Person Company. 30+ specialist agents running 24/7 on ClawBot + AgentOS, ~1B tokens/day in real production use. Teaches senior ICs how to install a crew without becoming a manager. 50,000+ students across Udemy and LinkedIn Learning. 18 AI products in 18 weeks to 5,200+ real users.
Previously at
Go deeper with a course
Ship a Real AI Product with a Self-Improving Agent, No Engineers Needed
Nathan Wang
Founder, Ipsion AI & Soul AI | 18 products in 18 weeks | PhD, ex-Apple & Meta