AI agents are set to become a company’s basic unit, says Lee Kai-fu

Lee’s vision reframes the firm as an orchestration layer. Instead of single employees owning discrete workflows, autonomous AI agents would execute tasks in finance, HR, legal and customer service, with managers coordinating agent swarms that operate 24/7. Because agents can be duplicated instantly and tuned for specific objectives, they introduce a form of elastic capacity that traditional hiring cannot match. In Lee’s framing, this is the mechanism behind the idea that future billion-dollar companies could scale with far leaner human headcounts.

The competitive race is already underway. Global and Chinese tech leaders are building agent platforms on top of large language models, adding memory, tool use and long-context reasoning so agents can plan and act with minimal supervision. Enterprises are piloting internal “copilots” that evolve toward autonomous agents as guardrails and monitoring mature. The early wins tend to be in repeatable, rules-heavy domains, KYC checks, claims triage, procurement prep, where accuracy can be measured and exceptions escalated to human reviewers.

Execution risks remain material, and Lee’s blueprint implicitly assumes strong governance. Companies will need rigorous audit trails, bias controls and fallback procedures to keep agent actions compliant and explainable. They will also need to retool job roles around supervision, escalation and system design rather than rote processing. If organizations can meet those standards while proving reliable productivity gains, the shift from human-centric workflows to agent-centric operating models could redefine how firms scale, compete and allocate talent over the next cycle.