In collaborationWORLDwithCapgeminiECONOMICFORUMAl Agents in Action:Foundations for Evaluationand GovernanceWHITEPAPERNOVFMRER2025
Contents456000ForewordExecutivesummaryIntroduction1EvolkingtechicalfoundationsofAagerts,11Thesaofweamearchtecturedf anAlacent1.2Communicationprotocols and interoperabilty1.3Cybersecurityconsiderations,2FoundationsforAlagertevaluationandgovernance2.1Classification2.2Evaluation2.3RiskassessmentConclusian|ContributorsEndnotesofitsMembersAIAgents in Action: Foundations forEvaluation and Govemance,
AlAgentsinAction:FoundationsforEvaluation and GovernanceForewordChiefExecutiveOfficer,CapgeminiInventtheExecutiveCommitteehroughthe AlGovernance Alliance, the WorldandCanaeminiareadvancincation withtheAlcommunityagentic future. If adopters start small,iteratecarefully and apply proportionate safeguardsuctivity andsescould autpaceoversight andigned incentives, emergent risks andalongside the practical steps needed to adopt andgents must bedeploy them satfely, responsibly and effectivelysponsibility to guidetheirent withcare.Throughnce, Al agents can be integrated in waysamplity human ngenulty, promote Inovationto evaluatethem and outliningoveral quality of lfe. This paper is aa,affering guidance to help eadlyhe complex and often unevenAIAgentsinActian:Foundations tr Ev
Executive summaryThis paper explores the emeraence of Alaqents, outlining their technical foundationsclassification, evaluation and governance tosupport sate and ettective adoptionThisreporthasbeen tailored mainly for adootersclassification that differentiates agents by their roleautorormy, authorty, predicabity and opeatoalleaders and practitioners seeking to integrate Alcontext.Thirdly, it suggests a progressive governanceagents into organizational workflows and services.apradh.that dredy corects evalato andWhileAlacents are aainina traction. there remnalinsdeployment environment.imited guidance on haw to design, test andovese them responsiby, Ths perans tohetpTogether,these elements guide adopters with afilthat gapby providing a structured foundation foconopaualbuepit ftrmong tom epetmenation,the sate and effective deployment of these systemsto deployment. The report highlights theimportancof aligning adoption with evaluation and govemancxsure that Alagents are successtulyeployed while trust, safety and accountabilityincluding their archit ectures, protocols and security considerations. Secondly, it offers a functionalAIAgentsinAction:Foundations forEvaluation andGovemance,
IntroductionAl agents are shifting from prototypes todeplovment. brinaina both transtormativeopportunities and novel governance challenges,Alagerntsaegaluly becoming embededintolay sthesedal-diven modes, paticulatyan increasing number of tasks, workflows andartficial inteligence (Al) andlarge languaguse cases that span cloud and edge computingAs the transition from protatyping todeploymenaccelerates, curent adoption remains concentratedamongearly adopters. According to arecent globanot,82%of organizations plan tointegrate agents within the next ane...