Artificial intelligence will fail to deliver broad economic returns unless corporations dismantle legacy job roles and rebuild their internal workflows around autonomous software agents, Baidu Inc. founder and Chief Executive Officer Robin Li said Thursday.
Speaking at the Baidu World 2026 conference in Beijing on Nov. 12, Li unveiled what he termed the "Three Principles of Organizational Evolution," arguing that the next phase of enterprise AI competition is shifting from raw model supply to structural business redesign.
"If artificial intelligence is the new electricity, then the intelligent organization is the new factory of the AI era, the place where productivity truly happens," Li said during his keynote address.
Li pointed to a growing contradiction across corporate AI deployments: while foundation models are improving rapidly and software agents are proliferating, isolated productivity gains by individual employees rarely translate into company-wide revenue growth or higher profit margins.
Demand for raw model capacity continues to surge. Baidu disclosed that its recently launched large model trading platform, Token Exchange, has attracted more than 100 supply providers during its first month of trial operations, surpassing 50 million daily calls and processing over 700 billion tokens per day.
However, Li warned that high token consumption does not automatically yield business returns. When employees use AI to draft documents, analyze data, or write code faster, those time savings are often lost in outdated cross-departmental hand-offs, manual approvals, and rigid decision-making hierarchies.
"AI has not generated the expected economic benefits because we have only replaced the electric motor without redesigning the factory," Li said, echoing recent research from management consultancies Boston Consulting Group and McKinsey that framed AI adoption primarily as an organizational challenge rather than a technical one.
To bridge the gap between AI tools and enterprise profitability, Li outlined three core principles for building an intelligent organization, starting with the need to break down traditional job boundaries. Conventional companies operate around fixed job descriptions that pass tasks sequentially between departments, whereas AI agents can take over information retrieval, analysis, content generation, and partial execution, shifting the basic unit of corporate operation from static roles to end-to-end tasks.
Li predicted that traditional positions will not disappear entirely, but will instead converge into two primary categories: "Builders," who create new products and systems, and "Sellers," who focus on communication, human connection, and building client trust. To support this shift, Baidu upgraded its general-purpose agent platform, Baidu Dazi, into a unified gateway connecting its full suite of specialized agents, while opening its Harness runtime engine and Agent API so AI tools can embed directly into existing corporate software rather than sitting in isolated chat windows.
Li's second principle centers on institutionalizing tacit knowledge within the enterprise. Much of a company's competitive advantage resides in the unwritten intuition of veteran staff, including how they evaluate clients, spot operational risks, and troubleshoot complex problems, all of which vanish when employees depart. Baidu Dazi's custom agent service and upgraded Enterprise Collaboration Workspaces are designed to capture business logic, project context, and decision-making criteria as reusable digital "Skills" and workflows that remain permanently inside the organization.
The third principle calls for designing closed-loop operational processes, which Li summarized with the formula "Measurable plus Looping equals Continuous Evolution." This approach requires tying AI agents to concrete business indicators such as delivery cycles, service quality, revenue, and operating costs so algorithms can self-correct through iterative feedback. To advance this model, Baidu introduced upgrades at the conference to its self-evolving algorithmic agent, Baidu Famou 3.0, which continuously optimizes production operations, commercial strategy, and industrial research around real-time performance metrics and currently serves clients across 25 of China's 41 major industrial categories.
Baidu presented both internal metrics and external industrial deployments to demonstrate how the framework operates in practice.
Over the past year, the number of daily active AI agents inside Baidu grew from 5,000 to 37,000, while completed tasks jumped from 84,000 per week to more than 1 million. Employees have built 4,300 shared Workspaces and codified over 20,000 organizational Skills. Within the Baidu Dazi development team itself, AI agents generated 90 percent of the code and 50 percent of external communications materials across more than 160 product iterations over the past six months.
In Baidu's Digital Human Innovation Business Unit, collapsing boundaries among product, engineering, design, and testing allowed a recent university graduate to use AI tools to deliver a project in one day that previously required five distinct roles and at least 10 days.
Outside the company, China Southern Power Grid partnered with Baidu to deploy an agent system named Tianxuan Partner, which integrates historical fault records, switch signals, and grid topology with veteran engineers' diagnostic expertise. According to data released at the event, the system increased distribution grid fault analysis efficiency twelve-fold and has been rolled out across 14 prefecture-level cities and 112 counties in the southern Guangxi region.
To support large-scale agent deployments, Baidu AI Cloud upgraded its Full-Stack AI Cloud architecture to orchestrate computing power, models, data, and agents directly around business workflows. The company also unveiled its Tianchi 1,024-card computing supernode alongside its ERNIE 5.5 Flash Preview model and two specialized application models.
Li noted that transitioning into an intelligent organization unlocks opportunities for asset-light operations, stronger competitive moats, and new market discovery. Still, he cautioned that simply installing AI agents does not complete a company's transformation, stressing that chief executives must personally lead overhauls of data permissions, human review checkpoints, and employee performance evaluations.
