智能体组织：AI时代新范式的轮廓

People & Organizational Performance PracticeThe agentic organization: Contours of the nextparadigm for the AI era Companies are moving toward a new paradigm of humans working togetherwith virtual and physical AI agents to create value. We share lessons fromearly adopters—and what you can do next. This article is a collaborative effort by Alexander Sukharevsky, Alexis Krivkovich, Arne Gast, Arsen Storozhev,Dana Maor, Deepak Mahadevan, Lari Hämäläinen, and Sandra Durth, representing views from McKinsey’sPeople & Organizational Performance Practice, McKinsey Technology, and QuantumBlack, AI by McKinsey. AI is bringingthe largest organizational paradigm shift since the industrial and digital revolutions(see sidebar, “The evolution of operating models”). This new paradigm unites humans andAIagents—both virtual and physical—to work side by side at scale at near-zero marginal cost. Wecall it the agentic organization. McKinsey’s experience working with early adopters indicates that AI agents can unlocksignificant value. Organizations are beginning to deploy virtualAI agentsalong a spectrum ofincreasing complexity: from simple tools that augment existing activities to end-to-end workflowautomation to entire “AI-first” agentic systems. In parallel, physical AI agents are emerging.Companies are making strides in developing “bodies” for AI, such as smart devices, drones, self-driving vehicles, and early attempts athumanoid robots. These machines allow AI to interfacewith the physical world. The evolution of operating models mirrored industrial thinking, hard-codingbusiness processes into monolithicsystems supporting production andenterprise resource planning. Companiessoon shifted to modular digitalproductsand platforms, updated monthly or evendaily.3Speed requiredagile operatingmodelswith small, cross-functional teams,including new roles such as softwareengineers, experience designers, andproduct managers. Speed and customeraccess became the keys to companies’competitive advantage. Today, 5.8 percentof the US population is employed in techjobs,4with only 1.6 percent in agriculture5and 19.3 percent in the industrial sector.6 The promise of this new paradigm willdepend upon the continued growth ofAI’s capabilities, as well as other factorssuch as regulation. The length of tasksthat AI can reliably complete doubledapproximately every seven months since2019 and every four months since 2024,reaching roughly two hours as of thiswriting.7AI systems could potentiallycomplete four days of work withoutsupervision by 2027. This would be aphenomenally accelerated evolution—from an intern-level employee requiringconstant supervision to a mid-tenureemployee who can operate independentlyto, perhaps, a senior executive who canshape and drive strategies. In the agricultural eraprior to the1800s, operating models were simpleand centered around small teams ofcraftspeople and farmers. Eighty to90 percent of the global populationworked in agriculture.1 Next, in theindustrial era, people movedinto factories, and operating modelsshifted to functional hierarchies. Productswere designed for mass replicationby people and machines, with majorupgrades every three to ten years. Newroles emerged, such as factory workers,engineers, and shift supervisors. Bythe 1970s, 39 percent of people in theUnited States worked in the industrialsector, with just 4 percent in agriculture.2Efficient scaling drove companies’ growthand competitive advantage, andleanmanagementbecame a strategic tool. Now, theAI erais beginning to usher inthe newest evolution, revolutionizingknowledge work like the previous erasrevolutionized physical work, with theagentic organization bringing togetherhumans, AI agents, and machines in theworkplace of the future. Organizational paradigms do coexist. Butthe agentic organization may offer thekey for the leaders to gain a competitiveadvantage by building decentralizedoutcomes-focused agentic networks. As thedigital eralaunched in the 1990s,industrial-age maxims began crumblingwith the rise of computing. Early IT efforts The agentic organization will be built around five pillars of the enterprise: business model;operating model; governance; workforce, people, and culture; and technology and data(Exhibit 1). Imagine, for instance, the bank of tomorrow: When a customer wants to buy a house,a personal AI concierge activates a series of agentic workflows to serve the buyer. A real estateAI agent suggests properties, while a mortgage underwriting agent tailors offers based on thecustomer’s financial profile. Compliance agents ensure that the deal adheres to bank policies,and a contracting agent finalizes agreements before another agent fulfills the loan. All theseworkflows are overseen by an agentic team of human supervisors, mortgage experts, andAI-empowered frontline employees. In some cases, the bank could even extend its AI-poweredservices into furnishing, renovations, energy upgrades, and more. The bank becomes a networkof