Agents,robots,andus:SkillpartnershipsintheageofAI

Authors

LareinaYee

AnuMadgavkarSvenSmit

AlexisKrivkovichMichaelChui

MariaJesusRamirezDiegoCastresana

November2025

Confidentialandproprietary.Anyuseof

thismaterialwithoutspecificpermissionofMcKinsey&Companyisstrictlyprohibited.

Copyright?2025McKinsey&Company.Allrightsreserved.

Coverimage?XH4D/GettyImages.

Allinteriorimages?GettyImages.

McKinseyGlobalInstitute

TheMcKinseyGlobalInstitutewasestablishedin1990.Ourmissionistoprovideafactbaseto

aiddecisionmakingontheeconomicandbusinessissuesmostcriticaltotheworld’scompanies

andpolicyleaders.WebenefitfromthefullrangeofMcKinsey’sregional,sectoral,andfunctional

knowledge,skills,andexpertise,buteditorialdirectionanddecisionsaresolelytheresponsibilityofMGIdirectorsandpartners.

Ourresearchiscurrentlygroupedintofivemajorthemes:

—Productivityandprosperity:Creatingandharnessingtheworld’sassetsmostproductively

—Resourcesoftheworld:Building,powering,andfeedingtheworldsustainably

—Humanpotential:Maximizingandachievingthepotentialofhumantalent

—Globalconnections:Exploringhowflowsofgoods,services,people,capital,andideasshapeeconomies

—Technologiesandmarketsofthefuture:Discussingthenextbigarenasofvalueandcompetition

Weaimforindependentandfact-basedresearch.Noneofourworkiscommissionedorfundedbyanybusiness,government,orotherinstitution;weshareourresultspubliclyfreeofcharge;andweareentirelyfundedbythepartnersofMcKinsey.Whileweengagemultipledistinguishedexternaladviserstocontributetoourwork,theanalysespresentedinourpublicationsareMGI’salone,andanyerrorsareourown.

YoucanfindoutmoreaboutMGIandourresearchat

/mgi

.

MGIdirectors

SvenSmit(chair)

ChrisBradley

KweilinEllingrud

SylvainJohanssonNickLeung

OliviaWhite

LareinaYee

MGIpartners

MekalaKrishnanAnuMadgavkarJanMischke

JeongminSeong

Agents,robots,andus:SkillpartnershipsintheageofAI1

Agents,robots,andus:SkillpartnershipsintheageofAI2

Contents

Ataglance3

Introduction4

CHAPTER1

Theworkforceofthefuturewillbeapartnership

ofpeople,agents,androbots7

CHAPTER2

Humanskillswillevolve,notdisappear,as

peopleworkcloselywithAI21

CHAPTER3

Entireworkflowscanbereimaginedaround

people,agents,androbots35

CHAPTER4

Leadershipiscrucialasagentsandrobots

reshapeworkandtheeconomy52

Glossaryofterms55

Acknowledgments56

Endnotes5

7

Agents,robots,andus:SkillpartnershipsintheageofAI3

Ataglance

—Workinthefuturewillbeapartnershipbetweenpeople,agents,androbots—allpowered

byAI.Today’stechnologiescouldtheoreticallyautomatemorethanhalfofcurrentUSwork

hours.Thisreflectshowprofoundlyworkmaychange,butitisnotaforecastofjoblosses.

Adoptionwilltaketime.Asitunfolds,someroleswillshrink,othersgroworshift,whilenewonesemerge—withworkincreasinglycenteredoncollaborationbetweenhumansand

intelligentmachines.

—Mosthumanskillswillendure,thoughtheywillbeapplieddifferently.Morethan70percentoftheskillssoughtbyemployerstodayareusedinbothautomatableandnon-automatablework.Thisoverlapmeansmostskillsremainrelevant,buthowandwheretheyareusedwillevolve.

—OurnewSkillChangeIndexshowswhichskillswillbemostandleastexposedto

automationinthenextfiveyears.Digitalandinformation-processingskillscouldbemostaffected;thoserelatedtoassistingandcaringarelikelytochangetheleast.

—DemandforAIfluency—theabilitytouseandmanageAItools—hasgrownsevenfoldintwoyears,fasterthanforanyotherskillinUSjobpostings.Thesurgeisvisibleacrossindustriesandlikelymarksthebeginningofmuchbiggerchangesahead.

—By2030,about$2.9trillionofeconomicvaluecouldbeunlockedintheUnitedStates—if

organizationspreparetheirpeopleandredesignworkflows,ratherthanindividualtasks,aroundpeople,agents,androbotsworkingtogether.

Agents,robots,andus:SkillpartnershipsintheageofAI4

Introduction

Workinthefuturewillbeapartnershipbetweenpeople,agents,androbots—allpoweredby

artificialintelligence.WhilemuchofthecurrentpublicdebaterevolvesaroundwhetherAIwillleadtosweepingjoblosses,ourfocusisonhowitwillchangetheverybuildingblocksofwork—the

skillsthatunderpinproductivityandgrowth.Ourresearchsuggeststhatalthoughpeoplemay

beshiftedoutofsomeworkactivities,manyoftheirskillswillremainessential.TheywillalsobecentralinguidingandcollaboratingwithAI,achangethatisalreadyredefiningmanyrolesacrosstheeconomy.

Inthisresearch,weuse“agents”and“robots”asbroad,practicaltermstodescribeallmachines

thatcanautomatenonphysicalandphysicalwork,respectively.Manydifferenttechnologies

performthesefunctions,somebasedonAIandothersnot,withtheboundariesbetweenthem

fluidandchanging.Usingthetermsinthisexpansivewayletsusanalyzehowautomationreshapesworkoverall

.1

ThisreportbuildsonMcKinsey’slong-runningresearchonautomationandthefutureofwork.

Earlierstudiesexaminedindividualactivities,whilethisanalysisalsolooksathowAIwilltransformentireworkflowsandwhatthismeansforskills.Newformsofcollaborationareemerging,creatingskillpartnershipsbetweenpeopleandAIthatraisedemandforcomplementaryhumancapabilities.

AlthoughtheanalysisfocusesontheUnitedStates,manyofthepatternsitreveals—andtheirimplicationsforemployers,workers,andleaders—applybroadlytootheradvancedeconomies.

Wefindthatcurrentlydemonstratedtechnologiescould,intheory,automateactivitiesaccountingforabout57percentofUSworkhourstoday.

2

Thisestimatereflectsthetechnicalpotentialfor

changeinwhatpeopledo,notaforecastofjoblosses.Astechnologiestakeonmorecomplex

sequencesoftasks,peoplewillremainvitaltomakethemworkeffectivelyandtodowhatmachinescannot.Ourassessmentreflectstoday’scapabilities,whichwillcontinuetoevolve,andadoption

maytakedecades.

AIwillnotmakemosthumanskillsobsolete,butitwillchangehowtheyareused.Weestimatethatmorethan70percentoftoday’sskillscanbeappliedinbothautomatableandnon-automatable

work.WithAIhandlingmorecommontasks,peoplewillapplytheirskillsinnewcontexts.Workerswillspendlesstimepreparingdocumentsanddoingbasicresearch,forexample,andmoretimeframingquestionsandinterpretingresults.EmployersmayincreasinglyprizeskillsthataddvaluetoAI.

Tomeasurehowskillscouldevolve,wedevelopedaSkillChangeIndex(SCI),atime-weighted

measureofautomation’spotentialimpactoneachskillusedintoday’sworkforce.Nearlyevery

occupationwillexperienceskillshiftsby2030.Highlyspecialized,automatableskillssuchas

accountingandcodingcouldfacethegreatestdisruption,whileinterpersonalskillslikenegotiationandcoachingmaychangetheleast.Mostothers,includingwidelyapplicableskillssuchasproblemsolvingandcommunication,mayevolveaspartofagrowingpartnershipwithagentsandrobots.

Agents,robots,andus:SkillpartnershipsintheageofAI5

Employersarealreadyadjusting.DemandforAIfluency—theabilitytouseandmanageAItools—

hasjumpednearlysevenfoldintwoyears.TheneedfortechnicalAIskillsemployedtodevelop

andgovernAIsystemsisalsogrowing,thoughataslowerpace.Abouteightmillionpeopleinthe

UnitedStatesworkinoccupationswherejobpostingsalreadycallforatleastoneAI-relatedskill—afractionofwhatmaybeneededintheyearsahead.Demandisalsorisingforcomplementaryskillssuchasqualityassurance,processoptimization,andteaching,aswellasforsomephysicalskills

suchasnursingandelectricalwork.Incontrast,jobpostmentionsaredecliningforroutinewritingandresearch,bothareaswhereAIalreadyperformswell,althoughtheseskillsremainessentialformuchoftheworkforce.

Inourmidpointscenarioofautomationadoptionby2030,AI-poweredagentsandrobotscould

generateabout$2.9trillioninUSeconomicvalueperyear.

3

Capturingthismaydependlessonnewtechnologicalbreakthroughsthanonhoworganizationsredesignworkflows—especiallycomplex,

high-valueonesthatrelyonunstructureddata—andhowquicklyhumanskillsadapt.IntegratingAI

willnotbeasimpletechnologyrolloutbutareimaginingofworkitself—redesigningprocesses,roles,skills,culture,andmetricssopeople,agents,androbotscreatemorevaluetogether.

Leaderswillplayacentralroleinshapingthispartnership.ThemosteffectivewillengagedirectlywithAIratherthandelegating,investinthehumanskillsthatmattermost,andbalancegainswithresponsibility,safety,andtrust.Theoutcomesforfirms,workers,andcommunitieswillultimatelydependonhoworganizationsandinstitutionsworktogethertopreparepeopleforthejobsof

thefuture.

Agents,robots,andus:SkillpartnershipsintheageofAI7

CHAPTER1

Theworkforceofthefuturewillbeapartnershipof

people,agents,androbots

AIisredefiningtheboundariesofworkandunlockingnewpotentialforproductivity.

4

Workwillbereconfiguredasapartnershipbetweenpeople,agents,androbots.

5

AIhasmadeagentsandrobotsmoreautonomousandcapable

Formuchofthepastcentury,machineshavebeenbuilttofollowrules.Robotsexecutedphysicalroutineslikeassemblingpartswhilesoftwareautomatedpredictableclericalandanalyticaltasks.

Bothtypesofmachinesoperatedinapredeterminedway;theydidwhattheywereprogrammedtodo,andlittlemore.TheriseofAIhasbeguntochangethatandtobroadenthescopeofwhatautomationcando.(Seesidebar“Howtechnologyisadvancing.”)

AIagentsandrobots—machinesthatperformcognitiveandphysicalwork,respectively—are

becomingmorecapableastheylearnfromvastdatasets.Thisenablesthemtosimulatereasoningandtorespondtoawiderrangeofinputs,includingnaturallanguage,andadapttodifferent

contextsinsteadofsimplyfollowingpresetrules.

Weestimatethattoday’stechnologycould,intheory,automateabout57percentofcurrentUSworkhours(Exhibit1).Thisfigurecomparesthecapabilitiesofexistingtechnologies,includingthose

demonstratedinalab,withthelevelofhumanproficiencyrequiredfordifferentworktasks.

6

Astechnologyadvances,thepicturewillcontinuetoevolveandshouldbeupdatedregularly.

Actualadoptiondependsonmorethantechnicalcapability.Factorsincludingpolicychoices,laborcosts,implementationexpenses,anddevelopmenttimeallinfluencewhenandwhereautomationisdeployed.Electricitytookmorethan30yearstospread,andindustrialroboticsfolloweda

similarmulti-decadepath.Asrecentlyas2023,onlyaboutoneinfivecompaniesranmostoftheir

applicationsinthecloud,despitethetechnologybeingwidelyavailablesincethemid-2000s.7

(Seethetechnicalappendixfordetails.)

Inthischapter,wefocusontechnicalautomationpotential—mappingthefrontierofwhattoday’s

technologiescandoandidentifyingthetypesofworkthatcouldbemostaffectedintheyearsahead.

Agents,robots,andus:SkillpartnershipsintheageofAI8

Sidebar

Howtechnologyisadvancing

Rapidadvancesinmodelreasoning

andcomputingpowerhavedramatically

acceleratedAI’sprogress.AImodelstrainedtosimulatereasoningareintegrating

disparatestructuredandunstructureddatasources,executingmultistepprocesses,

andabletomatchhumanperformanceinhighschoolanduniversitystandardized

examsacrossmultiplesubjects.Atthe

sametime,theadventandenhancementofgraphicsprocessingunits(GPU)and

tensorprocessingunits(TPU)aremakingmodeltrainingandinferencefaster,

cheaper,andmoreenergyefficient.AIhasalsobecomemultimodal,abletoingest

andgeneratetext,audio,images,and

video,anditisincreasinglyinteroperable

acrosstoolsandplatforms.Forexample,ModelContextProtocolandAgent2Agentareprotocolsthatallowteamsofagents

tocommunicate.Importantchallenges

remain,however,particularlyregarding

hallucinations,transparency,and

explainability,whicharekeytoensuring

safetyandavoidingunwantedbias.1The

underlyinginfrastructuretosupportAIisalsoadvancingquicklyfromGPUandTPUtotherapidbuild-outofAIdatacenters,

andnewtechniquestousetraditionalandalternativesourcesofenergy.

AI-poweredagentsasteammates

DevelopmentsinAIaretransforming

agentsfrompassiveassistantsinto

“virtualcoworkers,”withimproving

cognitivecapabilitiesthatcanincreasinglyautonomouslyplanandexecutecomplex

tasksinworkflows.2AIagentsarebeginning

tocarryoutmultistepprocessessuchas

interactingwithcustomers,processing

transactions,andcoordinatingfollow-up

actions.Thismarksafundamentalstep

towardAI-drivenoperations,wherepeople

andAI-poweredagentscollaborateasateamtodeliverresultsmorequicklyandefficiently.3

AI-poweredrobotsarebecomingmorecapable

Anewgenerationofgeneral-purposerobotsisemerging.PoweredbyAI,theyintegrate

spatialperception,reasoning,andaction

toperformcomplexphysicalactivitiessuchasoperatinginunstructuredenvironments,followingverbalinstructions,andexecutingvariationsontasksforwhichtheywerenotexplicitlytrained.TechnologicaladvancesinroboticsextendbeyondAItoinclude

improvementsindexterity,sensing,andedgecomputing.4

1IvanSolovyevandShresthaBasuMallick,“Gemini2.0:Levelupyourappswithreal-timemultimodalinteractions,”Google,December2024.

2McKinseytechnologytrendsoutlook2025,McKinsey,July2025.

3MarcBenioff,“Howtheriseofnewdigitalworkerswillleadtoanunlimitedage,”Time,November25,2024.

4“Aleapinautomation:Thenewtechnologybehindgeneral-purposerobots,”McKinsey,July2025.

AIcanhaveanimpactonalltypesofwork

Wedistinguishbetweenphysicalandnonphysicalwork.Robotsareneededtoautomatetheformer,agentsthelatter.Notallautomationrequiresagentsorrobotsinthenarrowtechnicalsenseofthoseterms,butweusethembroadlytocapturethefullrangeoftechnologiesthatautomatework.

Nonphysicalworkaccountsforabouttwo-thirdsofUSworkhours.Roughlyone-thirdofthose

hoursdrawonsocialandemotionalskillsthatmostlyremainbeyondAI’sreach,whiletherestinvolvetasks—suchasreasoningandinformationprocessing—thatarebettersuitedtoautomation.Thesemoreautomatableactivitiesrepresentabout40percentoftotalUSwagesandspanrolesinfields

fromeducationandhealthcaretobusinessandlegal(Exhibit1).

Thenear-terminfluenceofautomationonphysicalworkmaybenarrower.Activitiesthatrequirephysicalaswellascognitivecapabilitiesaccountforabout35percentofcurrentUSworkhours.

Robotshavemademajorprogress,butmostphysicalworkstilldemandsfinemotorskills,dexterity,andsituationalawarenessthattechnologycannotyetreplicatereliably(seesidebar“Robotsintheworkplace’’).

Exhibit1

Two-thirdsofUSworkhoursrequireonlynonphysicalcapabilities.

DistributionofphysicalandnonphysicalworkintheUS,byoccupationgroup

Capabilitiesrequired:1■PhysicalNonphysicalShareofworkthatrequiressocialandemotionalcapabilities

Shareof

workforce,%

Shareofworkhours,%

Occupationgroup

PhysicalNonphysical

705030101030507090

3

Buildingandgrounds

cleaningandmaintenance

4

Installation,maintenance,andrepair

5

Healthcaresupport

9

Transportation

andmaterialmoving

4

Constructionandextraction

8

Foodpreparationandservingrelated

1

Farming,?shing,andforestry

5

Production

2

Personalcareandservice

2

Protectiveservice

7

Management

6

Healthcarepractitionersandtechnical

8

Salesandrelated

1

Life,physical,

andsocialscience

1

Arts,design,entertainment,sports,andmedia

Architecture

2

andengineering

6

Educationalinstructionandlibrary

Agents,robots,andus:SkillpartnershipsintheageofAI9

Exhibit1(continued)

Two-thirdsofUSworkhoursrequireonlynonphysicalcapabilities.

DistributionofphysicalandnonphysicalworkintheUS,byoccupationgroup

Capabilitiesrequired:1■PhysicalNonphysicalShareofworkthatrequiressocialandemotionalcapabilities

Shareofworkhours,%

Shareof

workforce,%

Occupationgroup

PhysicalNonphysical

705030101030507090

O代ceand

12

administrativesupport

2

Community

andsocialservice

3

Computer

andmathematical

Businessand

7

?nancialoperations

1

Legal

TotalUSworkforce

10

30

50

50

10

70

70

30

90

1Allworkrequirescognitivecapabilities.Bothphysicalandnonphysicalworkmayalsorequiresocialandemotionalcapabilities.

Source:Lightcast;USBureauofLaborStatistics(2024);McKinseyGlobalInstituteanalysis

McKinsey&Company

Evenso,theeffectscouldbesignificantforsomeworkers.Physicaltasksmakeupmorethanhalf

ofworkinghoursforabout40percentoftheUSworkforce,includingdrivers,constructionworkers,cooks,andhealthcareaides.Advancesinroboticsareexpectedtochangeoccupationsinareas

likeproductionandfoodpreparation,includingsomelower-wageroles.Robotsmayalsocontinuetoperformworkthatishazardousorotherwiseunfeasibleforpeople,suchasunderwatertasks,search-and-rescue,andinspectionsofdangerousenvironments.

Agents,robots,andus:SkillpartnershipsintheageofAI10

Agents,robots,andus:SkillpartnershipsintheageofAI11

Sidebar

Robotsintheworkplace

Robotshavebeenaroundfordecades,butadvancesinAIaregivingthemcapabilitiesonceconsideredbeyondthereachofautomation.ThisprogressisbeingdrivenbyembodiedAI—theintegrationofintelligenceandphysicalitythatenablesrobotstoperceive,reason,and

actincreasinglyautonomously.

Robotstodaytakemanyforms,depending

ontheirapplication.Theyrangefrom

autonomousvehiclesthatnavigateroads

todronesusedforinspectionordeliveryto

disk-shapedmachineswithwheelsthatcleanfloorsormovegoodsinwarehouses.Typicaldeliveryrobotsareroughlycubeshaped,

whilequadrupedrobotsthatresembleanimalscannavigateroughterrain.

Amongthese,humanoidrobotscontinuetocapturetheimaginationwiththeirrelatableappearance,fuelinggrowinginterest,new

marketentrants,significantinvestment,

andwidespreadpublicfascinationthroughvideosshowcasingtheircapabilities.1In

principle,humanoidsofferpracticalphysicaladvantages.Theycanoperateinphysical

spacesdesignedforpeople,reducingtheneedforcostlyreconfiguration.2

Yetmajorhurdlesremain.Chiefamong

themaredexterityandmobility,requiring

advancesinactuators,mechanicalrange,

andsensorimotorcontrol.Safetyisanotherbarriertoscale,particularlywhenAI

modelsareemployedtocontrolrobotsin

thepresenceofhumans,demandingboth

regulatoryclarityandtechnicalprogressincollisionavoidance,malfunctionprevention,

cybersecurity,andtransparencyinAI

decision-making.Powerisalsoalimitation:Mosthumanoidscanoperateuntethered

foronlytwotofourhourspercharge.Evenifperformanceimproves,affordabilitymaybedifficulttoachieve—per-unitcostsof

advanced,safemodelswouldneedtofallfromtoday’s$150,000–$500,000rangeintheUnitedStatestoroughly$20,000–$50,000toenablelarge-scaleadoption.3

Massadoptionofhumanoidrobots

inworkplaceshingesonovercoming

thesechallenges,buttheinvestment

andexperimentationnowunderwayare

advancingtheentirefieldandheighteningawarenessofpotentialapplications.

Meanwhile,nonhumanoiddesignswillcontinuetoproliferate,growingfastinvolumeandvariety.

1TheHumanoid100:Mappingthehumanoidrobotvaluechain,MorganStanley,February2025.

2“WillembodiedAIcreateroboticcoworkers?”McKinsey,June2025.

3“Humanoidrobots:Crossingthechasmfromconcepttocommercialreality,”McKinsey,October2025.

AI-poweredautomationwillchangework,butpeopleremainindispensable

Atcurrentlevelsofcapability,agentscouldperformtasksthatoccupy44percentofUSworkhourstoday,androbots13percent(Exhibit2).

8

Thismeansthatautomationcould,intheory,takeonamajorityoftheworknowdonebypeopleintheUnitedStates.Thatdoesnotmeanhalfofalljobswoulddisappear;manywouldchangeasspecifictasksareautomated,shiftingwhatpeopledoratherthaneliminatingtheworkitself.

Inaddition,workthatdrawsheavilyonsocialandemotionalskillsremainslargelybeyondthereachofautomationevenunderafull-adoptionscenario.Thisisbecausemanytasksrequirereal-time

awarenesssuchasateacherreadingastudent’sexpressionorasalespersonsensingwhena

clientislosinginterest.Peoplealsoprovideoversight,qualitycontrol,andthehumanpresencethatcustomers,students,andpatientsoftenprefer.

Extendingautomationfurtherwouldrequiretechnologiesthatcanmatcharangeofhuman

capabilitiescurrentlyunmatched.Agentswouldneedtointerpretintentionandemotion.Robotswouldneedtomasterfinemotorcontrol,suchasgraspingdelicateobjectsormanipulating

instrumentsinsurgery.

Agents,robots,andus:SkillpartnershipsintheageofAI12

Exhibit2

People,agents,androbotscouldallplaysigni?cantrolesintheworkforceofthefuture.

21

22

13

44

Workhourscoveredby:PeopleAgentsRobots

Shareoftotal

hoursthatrequire

socialandemotionalcapabilities,%

DistributionofworkhoursintheUS,bytechnicalautomationpotential,2024,%

Activitiesrequiringnonphysicalcapabilitiesonly165%oftotalhours

Activitiesrequiring

physicalcapabilities135%oftotalhours

8

Workthat

isnotautomatable

43%oftotalhours

15

Workthat

isautomatable

57%oftotalhours

8

1

Note:Technicalautomationpotentialshownisthelatescenarioofexpertestimates.TheearlyscenariooftechnicalautomationpotentialintheUSis65%of

currentworkhours.Inthisresearch,weuse“agents”and“robots”asbroad,practicaltermstodescribeallmachinesthatcanautomatenonphysicalandphysicalwork,respectively.Manydiferenttechnologiesperformthesefunctions,somebasedonAIandothersnot,withtheboundariesbetweenthemluidandchanging.Usingthetermsinthisinclusivewayletsusanalyzehowautomationreshapesworkoverall.

1Allworkrequirescognitivecapabilities.Bothphysicalandnonphysicalworkmayalsorequiresocialandemotionalcapabilities.

Source:USBureauofLaborStatistics;O*NET;CurrentPopulationSurvey,USCensusBureau;McKinseyGlobalInstituteanalysis

McKinsey&Company

Astechnologyadvances,theworkrequiringpeoplewillalsochange.Someroleswillshrink,othersexpandorshiftfocus,andnewoneswillbecreated.Recentdevelopmentsinradiologyillustrate

thisdynamic.Between2017and2024,radiologistemploymentgrewbyabout3percentperyeardespiterapidadvancesinAI,anditisexpectedtocontinuegrowing.

9

AIaugmentedradiologists’work,improvingaccuracyandefficiencywhileenablingdoctorstofocusoncomplexdecision-

makingandpatientcare.

10

TheMayoClinic,forexample,hasexpandeditsradiologystaffbymorethan50percentsince2016whiledeployinghundredsofAImodelstosupportimageanalysis

.11

AIisalsocreatingothernewtypesofworkandroles.Softwareengineersarebuildingandrefiningagents,whiledesignersandcreatorsareusinggenerativetoolstoproducenewcontent.

Agents,robots,andus:SkillpartnershipsintheageofAI13

Sidebar

FramingthejobsdebateasAIreshapeswork

TheimpactofAIonjobsremainsuncertain.Whilemanystudiesattempttoestimate

potentialjobgainsorlosses,ourfocusisonhowtechnologyischangingthecontentofworkandtheskillspeopleneed,ratherthanonhowmanyjobsmayultimatelybegainedorlost.

Historysuggeststhatalthoughtechnology

hasdisplacedworkersintheshortterm,theeconomyhasgeneratedadditionaldemand

forlabor,includingnewrolesandindustries,overtime.ThebreadthofAI’scapabilities—itsreachintoreasoning,communication,and

judgment—hasheightenedconcernabout

thefutureofwork.Toframethedebate,we

explorewhatthecurrentresearchcanand

cannottellusthroughfourguidingquestions.

HowcloseareAIagentsandrobotsto

matchingalleconomicallyrelevanthumancapabilities?

AIisencroachingonworkonceconsideredbeyondautomation,extendinginto

reasoning,communication,andjudgment—skillsthatunderpinmostjobsinthemoderneconomy.1Despitetheseadvances,AIstill

lacksmanydistinctlyhumanabilities,leavingampleroomforhumanlabortothrive.To

matchpeopleentirely,machineswouldneedtogeneralizeandadaptacrosscontexts,

demonstrateadvancedfinemotorskills,

coordinatereliablyatscale,exercisesocial

andmoraljudgment,andtakeresponsibilityforoutcomes,allatacceptablecostand

risk.2Andbeyondthetechnicalautomation

potential,actualadoptionratesdependon

factorssuchassolutiontimelines,technologyversuslaborcosts,andthespeedatwhich

technologiesdiffusefromintroductiontowidespreaduse.

WillamoreAI-centriceconomycreateenoughjobs?

TheUSeconomyhascreatedtensofmillionsofjobsthiscentury,andprojectionsfrom

theUSBureauofLaborStatisticsand

WorldEconomicForumpointtocontinued

employmentgrowthoverthenextfive

totenyears.3Akeyissueiswhethernew

jobswillcomequicklyenough,andin

sufficientnumbers,toabsorbjobsthatare

displaced—andwhetherthosejobswill

havesimilarconditions.Whilethisisbeyondthescopeofouranalysis,pasteconomic

transformations—fromtheIndustrial

Revolutiontotheriseoftheinternet—offer

clues:Technologyhasofteneliminated

jobs,sometimesmassivelyandsometimes

depressingwagesincertainareas,buthas

ultimatelycatalyzednewindustriesandrolesovertime.4

EarlyevidencesuggeststhatAImayfollowthatfamiliartrajectory.Hiringhasreportedlyslowedforentry-levelpro