从即时到目的：用代理AI解锁商业价值

How to automate, optimize, and scale agentic AI Introduction In the enterprise conversation about AI, most of the attention –and the hype – has focused on generative AI and large languagemodels (LLMs): the type of AI exemplified by ChatGPT and its In this paper, we explore the technical prerequisites for success,the most common enterprise hurdles, and how the right platformapproach can address data fragmentation, unpredictable costs,and time-to-value. We also offer a practical checklist for the ideal Agentic AI is changing the conversation fast. According toCapgemini’s recent research,193% of leaders believe successfully deploying agentic AI at scale in the next 12 months will delivercompetitive advantage. Indeed, the pace of adoption was so rapidthat researchers felt the need to double-check with the majority Driving this adoption in the near term are process-intensivedomains such as customer service, IT, sales, operations, andresearch and development. As organizations seek to unlock newlevels of productivity, efficiency, and growth, the value of agentic However, realizing this value can be challenging. Powerful modelsare the foundation, but an effective agentic AI solution must alsobe built for ease, connectivity, and trust. This requires systemdesign that prioritizes seamless integration, clear governanceframeworks, and reusable, composable architecture, as well as What is agentic AI? Agentic AI is an autonomous system that sets its own goals, makesdecisions, and takes action to achieve them without any specificdirective instruction to do so. It consists of multiple differenttypes of intelligent agents acting together to gather, understand, The intelligent agents can include robotic process automation(RPA), machine learning models, and generative AI agents that cangenerate, validate, and summarize content as part of a broadertask. For example, Snowflake Cortex Agents are generative AIagents that help retrieve data insights from complex structured of organizations piloting or implementing AI agentsof business processes operating with semi to fullautonomy by 2028billion projected economic value generated by AIagents by 2028 across 14 countries37%$45025% Access to unified, AI-ready data:To get the full picture, AI agentsneed seamless access to complete, consistent, and connecteddata from all relevant sources. Agentic AI cannot act intelligently Self-optimizing AI pipelines:Agentic AI learns, adapts, andscales. Self-optimization makes that possible by turning feedbackinto improvement, which keeps independent agents safe, Real-time data processing:AI agents act in the moment. Theyneed real-time data to understand what is happening now and The critical features of Automated data governance:AI agents have to play by the rules.Strong governance and security frameworks ensure operational Composable AI workflows:In an agentic AI system, no agentworks alone. Composable workflows let businesses mix, match,and reuse AI components. It’s the key to real scalability and By its very nature, agentic AI can be deployed in a wide range of use cases.Whether an organization is implementing it to personalize the retail experience,accelerate clinical trials, or detect and prevent financial fraud, most successful As they prepare their organization and consider solutions and partners, businessesand tech leaders who are considering agentic AI should therefore look for critical Optimized compute power:AI agents process large volumesof data and make fast decisions. It’s compute-intensive work. Three industry hotspotsfor agentic AI value 1. Telecommunications Detecting and self-healing network faults: AI agents monitor,diagnose, and resolve network issues in real-time. Managing customer lifecycles: A hyper-personalized customerexperience, based on real-time activity, behavior, and account Monitoring regulatory compliance: AI agents continually scanbroadcast content and user interactions to flag or automatically Optimizing production lines: AI agents monitor demandfluctuations, equipment performance, and material constraints, Enhancing personal shopping: AI agents assist individualcustomers across multiple sessions and channels, curate Managing campaigns: Agentic AI designs and launches marketingcampaigns based on live performance data, then optimizes them Tracking environmental performance: AI agents collect, analyze,and report relevant data across supply chains for real-time Optimizing returns and refunds: AI agents handle post-purchasetasks, such as processing refunds, offering exchanges, flagging Maintaining equipment predictively: AI agents go beyondflagging anomalies to plan and schedule repairs, order parts, The five most common Just as there are common factors that are key to a successfuldeployment of agentic AI, so there are common barriers.Fragmented data, complex governance, cost inefficiencies, and A project facing one or more of the following red lights has amuch greater chance of failure