ADM在代理AI时代重新构想

Agentic execution forexperience-led outcomes The hidden ceilingof traditional andfirst-generationAI ADM Why ADM must beredefined now There is a growing disconnect between the valuedelivered by traditional application development andmaintenance (ADM) and the business requirements oftoday’s enterprises. Applications are increasingly coreto business success but are complex, and applicationsprawl creates inefficiencies and redundancies,security risks, and higher costs, especially in legacy-heavy environments. Constant change is now thenorm, fueled by artificial intelligence (AI)-acceleratedexpectations across the business. Yet ADM is stillmeasured primarily by cost, tickets, and service levelagreements (SLAs). Traditional ADM is optimized for efficiency rather thanadaptability. Operations react to issues as they ariseand depend heavily on people with the right expertiseto resolve them. Fragmented automation and toolinghave not been enough to deliver meaningful gains inproductivity or resilience. Early efforts to introduce AI and automation fell short.New tools were layered onto existing processes,generating incremental productivity gains but nosystemic change. Tickets were resolved faster, but theimpact on business outcomes remained limited. ADM therefore sits at the heart of a modern businessdilemma: how do you keep the lights on while stillfunding innovation? ADM is no longer a back-office efficiency play. AI makesa fundamentally new operating model possible, astep-change rather than a marginal improvement. Itbuilds the stable foundation that organizations needto meet today’s demands, while also delivering thetransformation required to keep pace with change. This focus on technical SLAs often failed to align withbusiness key performance indicators (KPIs), whichare what executives ultimately value. Because legacyADM models are not designed to support broadtransformation agendas, organizations struggledto keep pace with shifting market conditions andintensifying competition. Innovation remainedconstrained by high run costs and rigid delivery models. Without a new model for intelligence andaccountability, ADM remains a tactical functionrather than a driver of value creation. Moving from taskexecution tointelligent orchestration Agentic ADM introduces an agentic, AI-orchestratedmodel embedded across the application life cycle.Intelligence is designed in from the outset, ratherthan bolted on after the fact. This allows teams tomove beyond task execution toward stewardship ofoutcomes. The new operating model rests on three pillars: •An industrialized AI foundation•An Agentic ADM platform that is autonomous,orchestrated, and continuously learning•Experience Level Agreements (XLAs) that anchoraccountability to experience and value This adaptable approach also allows organizations toflexibly reuse and integrate existing AI capabilities.The result is more predictable transformation at scale,faster innovation, and lower run costs. The new ADM paradigm Capgemini’s new ADM AI SHIFT platform* - 3 layers Our promise Deployment ofCapgemini ADM AISHIFT platform in 12weeks within or outsideof your enterprise• •Flexibility to reuse andintegratesome of the AIlevers you have alreadydeployed in yourenterprise –Capgeminias the orchestratorofyour Agentic ADMapproach •No hidden costs withtheCapgemini ADM AISHIFT platform"already considered in ourbenefits commitments From automation toautonomy inday-to-day ADM In an ADM context, “agentic” refers to AI agents that canobserve, decide, and act across processes, continuouslylearning and adapting as they operate within theplatform. for higher-value activities. Greater stability and agilityallow businesses to respond more quickly to disruption,creating the capacity to modernize and innovate withfewer operational shocks. Autonomy delivers the greatest impact where work isrepeatable and standardized, for example, in knowledgecreation and reuse, or in the resolution of low-levelincidents with straightforward fixes that wouldotherwise require manual intervention. AI agents canpredict and resolve incidents, orchestrate change andrelease cycles, and maintain knowledge bases. We have moved fromhuman-led executionto human-supervisedautonomy. The shift is from human-led execution to human-supervised autonomy. Humans remain firmly in controlby design. Agentic AI is not a replacement for people,but a force multiplier that augments decision-makingwithin clearly defined guardrails and accountabilitystructures. The effects extend beyond IT operations. By reducingfriction, organizations free up time, talent, and capital Why experience,not activity,is thenew measure ofperformance XLAs do not replace SLAs, nor are they treated as anafterthought. Instead, they sit alongside SLAs andbusiness KPIs in a multilayer performance frameworkthat anchors outcome-based commitments. An agenticobservability layer combines application telemetry withbusiness context, enabling real-time issue d