Mcsey

Quarterly

QuantumBlack,AIbyMcKinsey

Thechangeagent:Goals,

decisions,andimplications

forCEOsintheagenticage

CompaniesarefeelingagenticAIgrowingpains.Here’swhatCEOscandoto

movepastthemandpositiontheircompaniestosucceed.

ThisarticleisacollaborativeeffortbyAlexSingla,AlexanderSukharevsky,LariH?m?l?inen,OanaCheta,Olli

Salo,PallavJain,RaghavRaghunathan,SandraDurth,StéphaneBout,andVitoDiLeo,representingviewsfrom

QuantumBlack,AIbyMcKinsey;McKinseyTechnology;andMcKinsey’sPeople&OrganizationalPerformanceand

Technology,Media&TelecommunicationsPractices.

October2025

ExecutivesarefondofquotinghockeygreatWayneGretzky,whoiscreditedwithsaying:“I

skatetowherethepuckisgoingtobe,notwhereithasbeen.”Thisissoundbusinessadviceatonelevel.ButthatpuckismovingawholelotfasterthanitusedtoasagenticAIrapidlyevolves.

ThecalltomovefastermayseemtonedeafasCEOsandtheirseniorteamsstruggletosee

bottom-linevaluefromearlygenAIinvestments.DevelopingandscalinggenAIusecaseshaveprovenfrustratinglychallenging.SomeexecutivesremainunconvincedthatAIagentswillhaveasignificantimpact—atleastintheshortterm—andhavesteppedbackfromtheirinvestments.1

AsCEOsnavigatetheuncertainty,itisworthacknowledgingboththepaceandpotentialscopeofthechangethatishappening.

AIagents

—softwaresystemsbuiltwithgenAIthathavethe

abilitytoplan,act,remember,andlearntoachievepredefinedoutcomesautonomously—are

evolvingquicklyand,astheymature,couldcompletelychangehowcompaniesarerunandhowthey

generatevalue

(seesidebar“KeytrendsshapinggenAIandagents”).Infact,this“troughofdisillusionment”period,asJohnLovelockofGartnerrecentlycalledit,2isanopportunityfor

executivestojumpaheadoftheircompetitors

.

HowCEOsmanagethischange

willdeterminehowwelltheycancapturethebenefits.AlthoughAIagentsareintheirinfancy,early

lessonsandexperiences

highlightfourmindsetsandactionsthatcanpositionCEOstoprosper:

—Reimaginewhat’spossible.MuchofthethinkingaroundagenticAItodayisstillfocusedonautomatingbasictasksoraugmentingknowledge.Therealwin,however,willcomefrom

muchbolderaspirationsofrearchitectedworkflowsandorganizationsbuiltaroundagent-firstsystems.

—Actwithurgencyandstartthelearning.TherapidrateofimprovementofgenAI

agentsmeansthatawait-and-seeapproachispotentiallyahigh-riskmove(seesidebar“BreakthroughsingenAIandagents”).Earlypracticallearningsareinvaluableinquicklybuildingacompetitiveadvantageasthetechnologymatures.

—Tacklescaleandlong-termcompetitivenessissuesnow.Criticaldecisionsaround

technology,trust,governance,whattobuyversuswhattobuild,capabilities,andtalentareimportanttodriveawidertransformation.Whileyouexperiment,startformingyourstrategyanddevelopingscalingcapabilitiesassoonaspossiblesinceexecutionwilltakelongerthanexpectedduetotalentscarcityandorganizationalcomplexity.

—Turneveryoneintoanagentleader.Asagentsandagenticsystemstakeovermoreofthe

executionalwork,everyoneintheorganizationwillneedtodevelopagentleadershipand

supervisionskills.Theexecutiveteamespeciallyneedstorolemodelandchampionlearningandtheevolutionoftheirworkhabits.

Whilemuchisstillunknown,buildingabusinessfortheagenticagewillrequirea

fundamental

rewiring

ofhowthebusinessoperates,innovates,andprotectssourcesofvaluecreation.Thisarticle,however,willfocusonafewofthemostimportantelementsan

enterpriseCEO

shouldaddressrelatedtovalue,scale,andtalent.Wewilloutlinewhatahypotheticaltwo-yearagenticjourneymightlooklike,whatkindsofdecisionsCEOsshouldconsider,andwhatthebig

implicationscouldbeforhowcompaniesoperate.

1“Gartnerpredictsover40%ofagenticAIprojectswillbecanceledbyendof2027,”Gartner,June25,2025.

2“WelcometotheAItroughofdisillusionment,”Economist,May21,2025.

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage2

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage3

KeytrendsshapinggenAIandagents

AIagentsarebecomingmorehuman-likeinthekindsoftaskstheycandoandthewaypeopleinteractwiththem.ThesefeaturesdemocratizeAIinawayprior

technologieshaventandunderscore

agentspotentialtoaffectabroadsetofactivities.TheincreasingpossibilitiesofgenAIthefoundationalcapabilitythatenablesAIagentsarefueledbyfour

mutuallyreinforcingtrends:

Anaccelerationininnovationpace.

Onlytwonew-frontierlargelanguagemodels(LLMs)wereannouncedin

2020;1by2025,thenumberisinthe

dozens,evenhundreds,dependingoncountingmethodologies.2Similarly,

thenumberofnewlarge-scaleAI

modelshasgrownby167percent

peryearsince2020.3Thelengthof

tasksAIagentscando(withatleast

a50percentsuccessrate)hasbeendoublingeverysevenmonths.4Atthetimeofwriting,ithasbeenreported

thatAnthropicsClaudeOpus4can

completealmostasmuchworkasa

humancaninaday,whileamultiagent

systemoutperformedasingle-

agentClaudeOpus4bymorethan

90percent.5

Largegrowthinspendand

investments.Thecomputeusedto

trainstate-of-the-artmodelshasbeengrowingroughlyfourtofivetimes

peryear.6Thetopthreehyperscalerscollectivelyplantoinvestmorethan$250billionin2025onAIanddata

centers,7andin2023,businessesspentabout$15billionongenAIsolutions,representingroughly

2percentoftheglobalenterprisesoftwaremarket.8

Sharpgainsinmodeltrainingand

inferencingefficiency.Breakthroughsinarchitectureandoptimizationhave

driventrainingcostsdownsignificantlyforagivencapability.Theinference

costsforChatGPT3.5droppedmore

than280timesbetweenNovember

2022andOctober2024.9Thecostpermillioninputtokens,forexample,has

decreasedabouttentimes,fromabout

$36.00inMarch2023toabout$3.50inAugust2024.10Forsomemodels,thecostislessthan$1.00.11

Breakthroughsinmodelandsystem

capabilities.Newreasoningmodels

deploy“testtimecompute”thinking

duringinference(“system-2thinking”);standardizedtool-callinginterfaces,

suchasAnthropicsModelContext

Protocol(MCP),letmodelsinvoke

enterpriseAPIssafely;vastlylarger

andmorepreciseshort-andlong-

termmemorystructuresimprove

boththerecallbreadthandprecision;multiagentorchestrationframeworks(forexample,LangGraph,AutoGen)

enablespecializedagentstodelegate,monitor,andreconciletheirwork;

andearlyagent-to-agentprotocols

(forexample,A2A,createdbyGoogleandrecentlydonatedtotheLinux

Foundationtomaintainasanopen-sourceproject)pointtoafuture

whereagentsautonomouslydiscoverpeers,negotiateroles,andexecute

workflows.

1The2024AIIndexreport,StanfordUniversity,Human-CenteredArtificialIntelligence,2024.

2Forexample,seeAnthonyCardillo,“Best44largelanguagemodels(LLMs)in2025,”ExplodingTopics,August28,2025.

3BasedonEpochAIdata,ascitedinMaryMeeker,JaySimons,DaegwonChae,andAlexanderKrey,Trends–ArtificialIntelligence,Bond,May2025.

4“MeasuringAIabilitytocompletelongtasks,”Metr,March19,2025.

5MichaelNu?ez,“AnthropicovertakesOpenAI:ClaudeOpus4codessevenhoursnonstop,setsrecordSWE-BenchscoreandreshapesenterpriseAI,”VB,May22,2025;“Howwebuiltourmulti-agentresearchsystem,”Anthropic,June13,2025.

6JaimeSevilleandEduRoldán,“TrainingcomputeoffrontierAImodelsgrowsby45xperyear,”EpochAI,May28,2024.

7DanRomanoff,“IsAIinvestmentpoisedforgrowth?Toppicksandpromisingapplicationsfor2025,”Morningstar,May27,2025.

8JeremySchneider,TejasShah,andJoshanCherianAbraham,“

NavigatingthegenerativeAIdisruptioninsoftware

,”McKinsey,June5,2024.

9The2025AIIndexreport,StanfordUniversity,Human-CenteredArtificialIntelligence,2025.

10“Largelanguagemodels:Performancevs.cost,”MentorCruise,January13,2025.

11“APIpricing,”O(jiān)penAI,accessedonSeptember19,2025.

Areagentsworthit?

ClaimsaboutthevalueofAIagentspermeatetheinternet.Butsincethetechnologyisstillso

new,thoseclaimsarehardtoverify.

Earlyimplementations,however,suggestthereissignificantvalueatstake.Ourexperiencewith

modernizingtechnologyestates

indicatesthatharnessingAIagentscanacceleratetimelines40to

50percentandreducecostsmorethan40percentwhilealsoimprovingthequalityoftheoutputs.3

3

“AIforITmodernization:Faster,cheaper,better

,”McKinsey,December2,2024.

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage4

Inanothercase,aleadinguniversalbankfacedamajorchallengeindeliveringahighvolumeofITprojectstodrivebusinessoutcomeswhilemanagingsignificanttechnicaldebtandashortageofskilleddevelopers.Startingwithasmallteamofthreeengineers,itbuiltatechmodernizationagentfactorywith100agentssupervisedbyjustfivehumans.Theseagents,underhuman

oversight,executedtheentiremodernizationlifecycle—fromreverseengineeringtodesigningandbuildingnewapplications—cuttingtimeandlaborcostsbymorethan50percent.

OurexperienceindicatesthattheinitialusesofAIagentstosupportpeopleandautomatetaskscandrive3to5percentinannualproductivityimprovementsatthecompanylevel.Asteamsof

AIagentsbecomecapableofcarryingoutmorecomplexworkflows,growthcouldincreasebyasmuchas10percentormore.

Knowyouragents:From‘a(chǎn)genticlabor’to‘a(chǎn)genticengine’

Executivesstilltendtohavefixedandlimitednotionsofwhatagentsareandwhattheycando.Thisconfusioncanmakeitdifficultforthemtounderstandwhatdecisionsrelatedtorisk,investment,resourceallocation,andchangeareneeded.

Whileit’stemptingtothinkofagentsinhumanterms,amoreobjectiveapproachistoconsiderthemassoftwaresystemsthatcandoaspectrumofincreasinglycomplextasks(exhibit).Inourexperience,thisapproachsharpensthethinkingaroundwhatsortsoforganizationalchangesarenecessary.

Agenticlabor:Agentsastoolstohelpwithexistingwork

Agentictoolscancontributetoexistingworkbyfacilitatingbasictasksundertakenbyindividualsandbyautomatingworkflows.

Individualaugmentation.Thesetoolshelpautomate,speedup,orimprovetasksthatpeopletypicallydo.Manyofthetasksarefamiliar—draftingresearchnotes,summarizingmeetings,generatingcode,conductingresearch,orproposingcontractclauses.

Thesetoolswilllikelybecome,andalreadyare,inareaslikeprogramming,a“costofdoingbusiness”similartoemployeesusingemailandspreadsheets.

Studiesshow20to30percenthigherpersonalthroughputandsometimesmuchhigher

numbersinsingle-taskareas.However,broadhorizontaldeploymentsofagentictoolingacrossthebusinessrarelytranslateintosignificantbusinessimpact.Furthermore,usagetendstotailoff,andretentiondropssignificantlybeyondleadusersformanytools.

Drivingbroadadoptionofpersonalsupportagentsrequiresfamiliarchangemanagement

investments,suchasembeddingthetoolsinstandardoperatingprocedures,integrating

theexpectedoutputsandusagemonitoringintoperformancemanagementsystems,giving

employeespropertrainingtousethetools,andcommunicatingandrolemodelingthebenefits.Atthesametime,leadershipwillneedtodeterminehowtocapturetheincreasedproductivitygivenitsspreadacrossseveralsmalltasks,oftendonethroughbudgetingandlarge-scale

efficiencyefforts.

Taskandworkflowautomation.Thesecondcategoryfocusesonautomatingexistingprocesses,workflows,andtasksintheorganization.Anagenticexecutionlayeressentiallysitsontopof

existingprocessesandsystems(withsmallchangestothem).

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage5

Exhibit

Agenticsystemscanworkonincreasinglycomplextasksdependingonhowwellcompaniescanovercomesomekeyconstraints.

Agenticsystemsbasedonincreasingtaskcomplexity

Agentic

system

Individual

augmentation

Taskandworklowautomation

Functionalagenticworklows

Cross-functionalagenticsystems

De?nition

Agentshelpauto-

mateandimprovebasictasks

Automationofexist-ing,low-complexityworklows

Agenticteamsworkingonredesignedworklows

Agent-drivensystemsworkon

complexworklowsacrossfunctions,

withhigh-leveldecisioningcapabilities

Primary

Individual

Coste代ciency,

Improvementsine代ciency,

Fasterproduction,lowercostper

bene?ts

productivity

speed,compliance

speed,andcustomersatisfac-

tion;greaterscale;revenueuplift

transaction,highervaluepercustomer,acceleratedcross-functionalprocessing

Key

Highlearning

Executioncapacity

Capacitytoreimagineprocess-

Organizationalandoperatingmodel

constraints

curve,continuedadoption,

signi?cant

changemanage-mentburden

toautomateexistingtasksatvolumeandathighquality

es,engineeringadvancestomanageagentteamsatscale

redesign

McKinsey&Company

Majortechnologyplayersareintroducingfirst-generationagenticproductswhileanexplosionofnewcompaniesisbringingsolutionstomanyfunctionaldomains(forexample,customercare,financialreportingandmonitoring,programming,productdevelopment,andprocurement).

Inourstudies,earlydeploymentshavedelivered20to40percentfastercycletimesorlower

handlingcostsforrepetitive,transactionalwork.Incontactcenters,certaintypesofcalls

(forexample,transactionalhandlingofbalancechecksandaddresschanges)arealmostfully

automated.Embeddingagentictoolsintoworkflowsandestablishingacontinuousimprovementapproachisacoreenabler,butsimplygivingtoolstouserswon’twork.

Theimpedimenttovalue,however,isthatthesemoredomain-specificusecasesoperatein

isolationandrelyonothersystemsandsignificanthumaninterventiontoperform.Furthermore,whilemodelcapabilityimproves,companiesstrugglewithexecutioncapacitytoautomate

existingtasksatvolumeandathighquality.

Agenticengine:Agent-nativeworkflowsandoperatingmodel

Emergingagenticsystems,drivenbybreakthroughsallowingforteamsofagentstowork

together,offerthemostpromisingopportunitiestogeneratemajorvalue.Capturingthisvalue,

however,requiresrethinkingandredesigningworkflowstobeagent-first,eitherwithinafunction(forexample,customercare)oracrossthem(forexample,leadtoorder).

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage6

Functionalagenticworkflows.Inthiscase,domain-specificworkflows(forexample,financialplanningandreporting)areredesignedtotakeadvantageofteamsofAIagentsandagentic

processes.Thatmeansrethinkingtaskorder,mergingtasks,accessingnewdatasources,anddevelopingnewprocesses,suchasearlysensingandresolutionofissuesbeforetheyemerge.

Agent-nativesystemscaneliminatethefrequenthandoffsandfragmentedactivitiesthathampermanycurrentprocessesbecauseteamsofagentsareorchestratedtoseamlesslyoperate.

Specialistvendorsinhorizontalandverticalsoftwarespacesarebuildingandimplementingfull-stack,agent-nativeapplicationsforareassuchascustomercare,finance,supplychainplanning,andsoftwaremodernization.Deployedcorrectly,thesesystemscutend-to-endcycletime,

improveresolutiontimes,anddriveupcustomersatisfaction.Forcallcenterperformance,for

example,theestimatedimpactcouldbeanautomatedhandlingof60to80percentofincomingrequestswithacomparableorbettercustomersatisfactionscorethanforcurrentsystems(forexample,interactivevoiceresponseplusfirst-linesupport).4

Suchsystemswillrequireacombinationofengineering(forexample,integratingprobabilistic

modelswithmoreclassical,deterministicsoftware)anddomainexpertisetobothbuildthe

multiagentsystemsandredesigntherelatedorganizationandoperatingmodelswithsufficient

humanoversight.Itwillbecriticaltoinstillintheseagentsgovernancerules(forexample,accessrights,decisionrights,andqualitygates)fortargetedworkflows(forexample,procuretopay,

vendorcontracting,suppliercommunication,andpolicymanagement)toensurethatsupervisinghumansaren’tquicklyoverwhelmed.

Cross-functionalagenticsystems.Theseagent-firstsystemsworkoncomplexworkflows(suchasend-to-endcustomerjourneys)acrossfunctionsandhavehigh-leveldecisioningcapabilities.Consider,forexample,24/7fieldserviceoperationsagentsthatdispatchtechnicians,

reschedulevisits,andorderpartsautonomously;aninsuranceteamthatadjudicatesclaims;a

mortgagethatisapprovedandunderwritteninseconds;orafinancialcyclewithagentshandlingeverythingfromannualplanningtomonthlyreporting.

Theseagenticsystemscancreatemultidimensionalvalueintermsof,forexample,fastertimetomarket,lowercostpertransaction,fasterissueresolution,andincreasedrevenuethroughbetteroffertargeting.Earlypilotsusingexistingtechnologiessawupto70to80percentreductionsincostpertransactionforcertainlabor-heavyprocesses.

Atthislevel,thekeyconstraintsaretiedtoorganizationalandoperating-modelissues.The

CEOandboardwillneedtobeintimatelyinvolvedtorearchitecttheoperatingmodel,includingleadershipandteamresponsibilitiesthathistoricallyhavelivedinsiloedcorporatefunctions.

Incrementalchangeswon’twork;thisleveloftransformationrequiresaclearbreakwithpastpractices.

Decisionstomakealonganagenticjourney—andsomebigimplicationstoconsider

TohelpCEOsvisualizethejourneyandsurfacesomecriticaldecisionsalongtheway,we’ve

laidoutahigh-level,hypotheticaltwo-tothree-yearroadmap.IthighlightscertainmarkerstoaimforandsomeofthekeydecisionsthatrequiretheCEOtobeactivelyinvolved.(Seesidebar“Howstart-upsbuiltaroundAIagentsarereshapingbusiness”forouttakesfromaninterview

4“

SeizingtheagenticAIadvantage

,”McKinsey,June13,2025.

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage7

withMagnusGrimeland,thecofounderandCEOofthetechventurecapitalandincubatorcompanyAntler.)

Thegoalsandtimelinespresentedintheroadmapareaggressive,andweacknowledgethat

muchwillchangeoverthetwo-yeartimeline.Inourexperience,however,itiscrucialforCEOstosetboldaspirationsandgoalstomotivatethebusinessandmovewithurgency.

Yearsoneandtwo:Settingthecourseandcreatingmomentum

Initialgoalsinyearoneshouldincludebuildingunderstanding,creatingmomentum,and

developingthefoundationssoAIagentscanworkatscale.Thefocusshouldbeondrivingdownoperatingcostsofexistingactivitiesinatargetedsetoffunctionsandoperations(considera

goalofupto10percentefficiencygains,forexample).Firstandforemost,though,thisphaseisaboutbreakingthroughtheinertia,movingwithpurpose,and

learning

.

Howstart-upsbuiltaroundAIagentsarereshapingbusiness

MagnusGrimeland,cofounderandCEOofventurecapitalfirmAntler,talkedwithMcKinseySeniorPartnerLariH?m?l?inenabouthowstart-upsaretakingadvantageofAIagents.Thefollowingisanedited

excerptofthatconversation.

LariH?m?l?inen:Doyouseeagentic-firstmodelsdisruptingbusinessmodels?

MagnusGrimeland:Absolutely.Some

companiesarejustbuildingmoreeffectivecoststructures,suchasautomatingthe

backofficeorbigpartsofthedevelopmentcycle.Butsomecompaniesarebeing

muchmoredisruptivebycompletely

removinghumansfromcorepartsofthe

businessmodel.Onelogisticscompanywebackedislookingtocompletelyrearchitectandreplacetheback-officefunctionsthatoptimizetheentiresupplychainaroundAIagents.Anotheroneislookingtoreplace

processengineersfromfactories.Therearetremendousopportunitiesacrossthevaluechain.

LariH?m?l?inen:Amongstart-ups

thatarebuildingaroundAIagentsandAIingeneral,howaretheybuildingupbusinessesdifferently?

MagnusGrimeland:That’sabroad

question,butafewthemeshaveemerged.Oneisthatwe’reseeingcompaniesgrow

fasterwithsignificantlyfewerpeople

thanwe’veeverseenbefore.Two,they

typicallyhavemuchflatterstructures,

withleadershipbeingmuchmoreactiveinproductdevelopmentandsales,andthe

wholeteamiscomposedofbuilders.Inthepast,whenstartingupalogisticscompany,forexample,you’dlookforpeoplewithtento15yearsofexperienceintheindustry.

WithAI,itwillbemoreandmoreaboutfindingthebesttechnologytalentandengineers.

Andanotheristhatstart-upsarereallykeenlyattunedtothespeedofchange.

Thatmeanstheythinkintermsofhow

tocontinuouslyupskillandreskilltheir

people,andtheyconstantlyscanthe

marketforbreakthroughsandpotential

competitors.AIisdevelopingataspeed

thatismuchfasterthanpreviouscycles,

andyouquicklybecomeirrelevantifyou’renotontopofnewcapabilities.

Wealsoseethisspeedandflexibility

playingout,ascostshavecomedownsignificantly.Inthepast,whenyoubuilt

ane-commercesite,youneededatech

backbone,whichneededalotofplanningandlongproductdevelopmentcycles.

Withgen-AI-firstcompanies,theproductcycleismuchfaster,andthereistheabilityforpeopletoswitchsystems.Youcan

buildontopofoneinfrastructure,then

switchovertoanotherrelativelyeasilyifit’sabetteroption.Thismeansitismucheasier,cheaper,andfastertobuildandexperiment.

LariH?m?l?inen:WhereisthevaluegoingtocomefromintheAIagentworld?

MagnusGrimeland:Whatvaluemeansandwhereitcomesfromhastoswitch.

Manyservicestodayareessentially

addingan“AIagent”buttonontotheir

existingservicesandthencharging30to40percentmoreforit.Thatmodelwon’t

workverywellforlong.Allcompanies

havetothinkaboutrebuildingAIfirst.Thatmeansrethinkinghowtoengagewiththeuser,howtoaccessinformation,andhowtoservecustomersdifferently.Thiswill

eventuallybeaboutreplacingservicesratherthanjustcreatingincrementalrevenuestreams.

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage8

Inyearsoneandtwo,lookforthefollowingagenticbusinessmarkers:

—Agent“fluency”growsquickly.BeingabletouseAIagentsproductivelyisarequirementforallworkers.Whilethevaluetothebusinessislow,buildingupthiscapabilityamong

employeesisthe“costofdoingbusiness.”Thegoalshouldbetohavemorethan25to50percentofemployeesworkingwithenterpriseagentsandAItoolsregularly.Allemployeesshouldbeinterrogatingdatathrough“chats”withagentsratherthanjustreadingreports.

—Agentsareautomatingabroadrangeofexistingprocesseswithfirst-generationtools.

Thisincludescriticalprocessessuchasfinancialfilingsandbroaderdocumentauthoring,

approvalsinexistingprocesses,andsoon.Concretebenefits,likesignificantlyfasterlead

timesandlowertransactioncosts,shouldbeclearlyevident.Forexample,intargetedcases,suchascorrectingsimple,well-structureddataqualityissues,agenticAIcanresolve90to

95percentoftheissues.

—Firstagenticsoftwaresystemsareintegratedintokeysystems.Keysystemsarechanging

theirinterfacestowardprompt-basedqueryingratherthanstaticcommands.Agentsarein

placeinkeysystemstoautomaticallycreateinsights,executetasks,andcoordinateactivities.Automatedcapabilitiessuchasplanningorinformationgatheringareincreasinglythenorm,

andsupportingsystems(suchasreporting)arechangingfast.

—Afront-runnerteamlaunchesalighthousetoreimagineacompleteend-to-endprocess.

Theteamdesignsa24-monthtargetstatevisionforacompleteprocess(forexample,

ordertocash,recordtoreport,automatedloanacceptanceandprovisioning)andstarts

byreleasingaseriesofminimumviableproductstotestandexpandcapabilities.Thegoalsshouldbebold.Foranorder-to-cashprocess,forexample,thegoalcouldbetoautomategreaterthan70percentoftransactionsacrossallchannels.

—Thedemandforcertainrolesmaylessenasproductivityincreases.AIagentshavereliablyandefficientlytakenoversimplecodingtasks,reducingtheburdenonsomeexistingroles.

Thelatestcodingagents(especiallyfront-endcodeexecution),forexample,significantlyincreaseproductivitybyasmuchas50to100percent.

EnablingthebusinesstohitthesemarkerswillrequireCEOstoaddresssomecorebusinessareas:

Architectingthetransformationforvalue.Toomuchofthecurrentfocusisonindividual

productivity;whileuseful,thisisn’twherethegreatestpoolsofvalueare.CEOsneedtoaspire

totransformationalvalue,whichwillcomefromrearchitectingandredesigningentireworkflowswithagents.

TheCEOwillneedtoensureteamsmovefromworkingonisolatedusecasestofocusingon

cross-functionalpriorityworkflows.Thiswillnecessitateanorganizationalshiftawayfrom

siloedAIteamstowardcross-functionalagenticteamsthatincludeAI,data,IT,technology,andfunctionalexpertsfromrelevantdomains.

Astheseteamswork,itwillbecrucialtoputapremiumonthelearningthatisgenerated.This

meansensuringthatenterprise-widelearningsarecentrallycapturedandreusedacrossthe

organization.Italsomeanscodifyinganagent-firstworkflowredesignplaybook,includingROI

criteria,multiagentorchestrationpatterns,techanddataintegrationbestpractices,controlsandevaluations,andwhenorwhennottoapplyagents.

Thechangeagent:Goals,decisions,andimplicationsforCEOsintheagenticage9

Toleadthiseffort,organizationswillneedacentralteam(“agenticfactory”)responsibleforidentifyingtheworkflows,managingtheredesign,andscalingtheredesign.

Scalingthetransformation.In2022,Amazon’sfounder,JeffBezos,mandatedthatdevelopers’codeincludeAPIsthatcouldbeexposedtothirdparties.Th