Kunlun Wanwei CEO says token-maxing has failed, and AI coding needs real engineering guardrails
昆仑万维集团 · wechat · 2026-07-23
At WAIC, Kunlun Wanwei CEO Fang Han argued that token consumption is a poor proxy for AI value, and that the real driver is an engineered harness like Claude Code: long context, tool use, code execution, and continuous feedback.
He said AI coding is now the clearest commercial token sink, but also creates technical debt, so code review, testing, permissions, sandboxes, and accountability must tighten. On org change, Kunlun has senior management alignment, uses AI coding tests for mid-level managers, deploys learning agents in work groups, and maintains shared team memory systems. Fang also said the company still believes models and compute are the long-term foundation, continues training its own models, and is betting on world models, video, music, and games as the next major growth area.
More from Companies & People
- Google maps the emerging AI economy in a new research blog post — swolpers · 2026-07-23
- OpenMed marks one year as healthcare stays the weak spot in open-source AI — MaziyarPanahi · 2026-07-23
- MIT joins 15 DOE Genesis Mission projects on AI for science, fusion, and quantum sensing — MIT News AI · 2026-07-23
- Meituan launches a 2026 embodied-AI robot contest with 550,000 RMB in prizes — 美团技术团队 · 2026-07-23
- Toby Walsh joins Chen Institute’s AIAS+ 2026 symposium in San Francisco — TobyWalsh · 2026-07-23
- Moonshot AI founder Yang chose entrepreneurship to avoid lifelong regret — nathanbenaich · 2026-07-23