Kai-Fu Lee says most corporate AI programs are ‘theater’ and predicts agents will reshape companies

Kai-Fu Lee has a message for the Fortune 500: your AI strategy is probably fake. In a Bloomberg interview, the CEO of 01.ai and chairman of Sinovation Ventures laid out a blunt assessment of corporate America’s AI adoption, calling most programs “theater” unless they’re actually moving the needle on earnings.

Coming from a former president of Google China who now runs one of Asia’s most prominent AI labs, the criticism carries weight. Lee’s broader thesis is that artificial intelligence is advancing faster than business leaders realize, and that the organizational chart of a typical company will look fundamentally different within five years.

The ‘theater’ problem and why CEOs are behind

Lee pointed to concrete capability improvements to underscore the gap between what AI can now do and what most executives seem to grasp. Tasks that once took 40 minutes can now be handled in four, he noted, a tenfold improvement in problem-solving efficiency.

His argument is that CEOs who treat AI as an incremental tool, something to bolt onto existing workflows, are missing the point entirely. The technology is moving toward replacing entire categories of work, not just speeding them up.

China’s open-weight play and the Global South opportunity

Lee also offered a detailed perspective on where Chinese AI development stands relative to US competitors. His assessment: China’s open-weight models trail American frontier models by roughly six to nine months in raw capability. But that gap, he argues, is less important than it appears.

Chinese open-weight models, such as Moonshot’s Kimi K3, are significantly cheaper to deploy and far more customizable than their proprietary Western counterparts. For companies in emerging markets, particularly across the Global South, this cost advantage matters enormously.

Beijing’s strategic embrace of open-weight models, where the model weights are publicly available for modification and deployment, creates a different competitive playbook than the closed-model approach favored by OpenAI and Anthropic.

‘Year One of Reasoning Agents’

Perhaps Lee’s boldest claim was his characterization of 2026 as “Year One of Reasoning Agents.” This is the moment, he argues, when AI agents stop being demos and start becoming the foundational units of how companies operate.

By 2031, Lee predicts that organizational charts will reflect this shift in dramatic fashion. AI agents won’t just assist human workers. They’ll replace many of them in operational roles, handling workflows that currently require teams of people.

Lee suggested that future career relevance will depend on skills like problem-solving, judgment, and the ability to direct AI systems effectively. Traditional coding and engineering become less central as AI handles more of the execution layer.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.