麦肯锡-为什么代理是生成式人工智能的下一个前沿（英）-2024.7-8页

July2024McKinseyDigitalPracticeWhyagentsarethenextfrontierofgenerativeAIBymovingfrominformationtoaction—thinkvirtualcoworkersabletocompletecomplexworkflows—thetechnologypromisesanewwaveofproductivityandinnovation.ByLareinaYee,MichaelChui,andRogerRobertswithStephenXuOverthepastcoupleofyears,theworldhasmarveledatthecapabilitiesandpossibilitiesunleashedbygenerativeAI(genAI).Foundationmodelssuchaslargelanguagemodels(LLMs)canperformimpressivefeats,extractinginsightsandgeneratingcontentacrossnumerousmediums,suchastext,audio,images,andvideo.ButthenextstageofgenAIislikelytobemoretransformative.Wearebeginninganevolutionfromknowledge-based,genAI–poweredtools—say,chatbotsthatanswerquestionsandgeneratecontent—togenAI–enabled“agents”thatusefoundationmodelstoexecutecomplex,multistepworkflowsacrossadigitalworld.Inshort,thetechnologyismovingfromthoughttoaction.Broadlyspeaking,“agentic”systemsrefertodigitalsystemsthatcanindependentlyinteractinadynamicworld.Whileversionsofthesesoftwaresystemshaveexistedforyears,thenatural-languagecapabilitiesofgenAIunveilnewpossibilities,enablingsystemsthatcanplantheiractions,useonlinetoolstocompletethosetasks,collaboratewithotheragentsandpeople,andlearntoimprovetheirperformance.GenAIagentseventuallycouldactasskilledvirtualcoworkers,workingwithhumansinaseamlessandnaturalmanner.Avirtu...