智能体AI优势：解锁下一代价值

KPMG. Make the Difference. AtKPMG,thefutureisagent-centric That includes preparing your people. The organizations that willlead will likely be those that help their workforces to adapt bytraining teams to work with agents, clarifying evolving roles and AI has reached another inflection point. Just as organizationsbegin to scale generative AI, a more transformative shift is already Agents aren’t just the next wave — they’re a step-change. Thesesystems will act independently, pursue goals and operate acrosscomplex workflows. That shift changes everything: how work gets And as agents gain more autonomy and influence, trust becomesmore critical. A scalable strategy isn’t only about what agents cando — it’s about ensuring that they operate reliably, transparently This paper breaks down what agentic AI really means: how todefine it, what it enables and where to begin. The perspectives aregrounded in the real world — drawn from the work we’re doing With all the hype, it’s easy to lose focus. Agents can be a criticalpart of AI strategy, but they aren’t the strategy. The leaderspulling ahead of the pack are moving fast and with clarity —staying anchored on the “no regrets” moves: aligning leadership, Steve ChaseGlobal Head of AI and DigitalInnovation, KPMG Internationaland Vice Chair – ArtificialIntelligence & Digital Innovation The time to prepare is now. Unlike cloud, which took nearly two decades to reshape enterpriseinfrastructure, this shift is playing out in months. Leadingorganizations aren’t waiting — they’re already investing, scaling and The shift won’t be easy. But it can be defining. The decisions thatleaders make today will likely reshape industries and redefine what This isn’t just a tech conversation — it’s a leadership one. Agentswill demand tighter alignment between business and technologyleaders than anything that’s come before. Success will require a As we continue our AI journey, we’ll continue to share whatwe learn — and welcome your ideas and perspectives as we Contents 11The spectrum of agents is evolving to meetcritical needs12From experimentation to implementation —Choosing the right agent for the task18Navigating the agentic AI obstacle course28How to accelerate your agentic-AI journey30How KPMG can help 04 05The power behind the promise of AI 06The size of the prize could be game-changing 07How much value could be at stake? 08 Four ways AI agents can unlock value WelcometotheageofagenticAI As organizations begin to scale thegame-changing capabilities of generativeAI (Gen AI), an even more advanced capabilityis emerging: the rise of AI agents. While still inits early days, agentic AI is evolving quickly and The number of digital workers may start out small buthas the potential to scale quickly, flexing up or down withdemand. By taking on routine tasks, digital colleaguesfree human workers to focus on higher-value, complex automate, and they may be an even better option forsome activities previously handled by earlier AI tools that The dawn of human/AI collaboration Agents’ ability to automate complex tasks is growing rapidly —currently doubling every three to seven months. In coding, theircapabilities have already grown from handling tasks that tookhumans seconds to complete to automating hours-long tasks.If this exponential growth continues, agents could be fourtimes more capable within six months and sixteen times more This multi-layered approach can unlock operationalperformance while also creating opportunities toreimagine entire business models. However, realizing this In simple terms,Gen AIbuilds “digital assistants” thathelp employees free up time to be more productive,effective and creative. Digital assistantsaugmenttheirhuman colleagues by providing information and guidance powerful in a year, enabling them to move from automating Agentic AIis the next leap forward — going beyondassistinghumans by fullyautomatingwhat humansdo and acting on their behalf to deliver the intendedbusiness outcomes. By integrating Gen AI tools, data Agents are also increasingly capable of communicating andcollaborating with each other, further amplifying their businessutility. By taking on a series of complex tasks as a group,“swarms”1of agents with different skills may eventually They’re already capable of taking on humantasks that were once considered too complex to ThepowerbehindthepromiseofAI Generative AI unlocked insights at scale. Agentic AI istaking the next step — turning insights into intelligent, reinforced feedback— all under appropriate levels ofhuman oversight both in the loop and on the loop. The tools and approaches they use are selectedand adapted dynamically, based on defined policies, The range of real-world applications is vast, fromtask-oriented agents that extract and compare dataor documents against standards, to fully autonomous KPMG International is pleased to present this collaborativereport built on pivotal research findings by KPMG in the