




版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
AI-AssistedCoding:
AugmentingSoftware
DevelopmentwithGenerativeAI
ExploringtheIntegrationofGenerativeAIinSoftwareEngineeringtoEnhanceCodingandTeamCollaboration
STNETNOC
04
05
07
07
08
08
09
09
11
12
13
14
GenerativeAIissoftwareengineering’slatestandgreatestevolution.
Bringingtomorrow’ssoftwareengineeringparadigmintoview:augmentedsoftwareteams
Howtomoveforward:Aprovenmethodforasoftwareengineeringtransformation
Assesstheorganization’smaturityanddeveloparoadmapwithclearobjectives.
Runreal-worldexperimentationsandmeasuretheimpactofGenerativeAI
DeployGenerativeAIforSoftwareEngineeringatscale
Capgemini’sexperienceinmeasuringGenerativeAIimpact
Measurementprotocolandreal-worldexperimentations:designedforactionableandreliableresults
Prerequisites&SuccessFactorsLessonslearned
Wecandrivesoftwareengineeringtransformationwithyou
Thedawningofaneweraforsoftwareengineering
2AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Disclaimer:Thispaperpresentsforward-lookingperspectivesgroundedinthecuΓΓe∩tla∩dscapeofGe∩eΓativeA/./tΓe卩ectstheΓapidadva∩ceme∩t—both
i∩teΓmsofouΓu∩deΓsta∩di∩ga∩dexpeΓie∩ceofGe∩eΓativeA/tech∩ologies./t
dΓawsi∩sightsfΓomexteΓ∩alΓeseaΓch,tech∩ologyleadeΓsaswellasouΓow∩
expeΓime∩tatio∩swithouΓteamsa∩dwithclie∩ts.Asalways,thefutuΓeΓemai∩su∩pΓedictable,butweca∩establisha∩dimagi∩elikelytΓajectoΓiesa∩doutlooks.
Softwareplaysacriticalroleinmodernbusiness,
whetherit’sintegratedintobusinessappsorproducts.Despiteitsubiquity,theultimatechallengeforsoftwareengineeringorganizationshasalwaysbeentorelease
qualitysoftwarefastenoughtokeepupwithconstantlyacceleratingdemand.
Overtheyears,productivityandqualitystandards
havebeenprogressivelyenhancedthroughthe
adoptionofnewmethodsandtechnologies.Agile
andcontinuousproduct-centricapproaches,software
lifecycleautomation,open-sourceecosystems,cloudnativeandcomposablesoftwarearchitectures,DevOpscontinuum,andlow-code/nocodedevelopmenthaveallbeenintroduced.Andeachhasbuiltontheadvancesofpastiterations.
Butdemandforquickertime-to-valueandbettercostefficiencyisstillaccelerating.Thisleavessoftware
organizationsstrugglingtodeliverattheexpectedpace,withtheexpectedquality,whilecontrollingtechnicaldebtandcosts.
3AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
GenerativeAIissoftwareengineering’slatestandgreatestevolution.
GenerativeAInowstirswonder
andexcitementacrossmany
applicationfields,notleast
softwareengineering.BillGatesiscallingitslatestadvancementthemostrevolutionarytechnologicalachievementinover40years,1
withthepotentialtodramaticallyimprovethewayorganizations
meetbusinessandITchallenges.
Theintersectionofbusiness
andtechnologywillbeattheheartofgenerativeAI’simpact.
Developmentsononeside
influencetheother.Forexample,
efficientsoftwareengineeringwithhighqualitycanreducetimeto
marketandthusprovidebusiness
valueearlier,givingleading
organizationsbothasoftwareandbusiness-orientededge.
OurCapgeminiResearch
Institutereportshowsthat61%
oforganizationsseeenabling
moreinnovativework,suchas
developingnewsoftwarefeaturesandservices,astheleadingbenefitgenerativeAI.Closebehindare
improvingsoftwarequality(49%)andincreasingproductivity(40%).Organizationsareutilizingtheseproductivitygainsoninnovativeworksuchasdevelopingnew
softwarefeatures(50%)and
upskilling(47%).Veryfewaimtoreduceheadcount(4%).2
AccordingtoanotherCapgeminiResearchInstitutesurveyacross
800organizations,67%of
executivesseethemostpotentialforgenerativeAIintheITfunctiontodriveinnovationandcreate
value3Moreover,accordingto
Forresterresearch,“Off-the-shelfandcustomAIsoftwarespend
willdoublefrom$33billionin
2021to$64billionin2025andwillgrow50%fasterthantheoverallsoftwaremarket,withanannualgrowthrateof18%.”4
Inbothlegacysoftware
modernizationandnewsoftwaredevelopmentcontexts,generativeAIgivesbacktimetosoftware
engineers.Theyarefreedup
tofocusonbusinessdemand,
softwarequality,security,andtheadvancedfeaturesrequiredby
newsoftware.
GenerativeAIalsohasapositive
impactonsoftwareprofessionals’jobsatisfaction.69%ofsenior
softwareprofessionalsand55%
ofjuniorsoftwareprofessionals
reporthighlevelsofsatisfaction
fromusinggenerativeAIfor
softwaredevelopment.78%
ofsoftwareprofessionalsare
optimisticaboutgenerativeAI’s
potentialtoenhancecollaborationbetweenbusinessandtechnologyteams5
GenerativeAIwillsupplement
bandwidth-usingthesame
capacity-forhigherproductivityandefficiency,alongwithfaster
timetomarket.Butonlyif
organizationsandtheiremployeescommittogettingonboard,
keepinginmindthatearlyadopterswilltakealeadingpositionin
thefield.
Inaddition,generativeAIpresentsanopportunitynotonlyto
transformbutalsotostandardizeandenhancethedeliveryof
software.Thebenefitsbrought
byusinggenerativeAIinsoftwareengineeringcanextendtoother
areasofthebusinesstocreateevenmorevalue,reduceITcostsand
minimizetechnicaldebt.
OrganizationsusinggenerativeAIhaveseen7%–18%productivity
improvement6inthesoftware
engineeringfunction,compared
totheirinitialestimates.Creatingliteratureanddocumentation,
andwritingcodeandscriptsshowthegreatesttimesavingwith35%maximumand10%average,and
34%maximumand9%average
respectively.7GenerativeAIalso
hasapositiveimpactonsoftwareprofessionals’jobsatisfaction.69%ofseniorsoftwareprofessionals
and55%ofjuniorsoftware
professionalsreporthighlevelsofsatisfactionfromusinggenerativeAIforsoftware.78%ofsoftwareprofessionalsareoptimistic
aboutgenerativeAI’spotentialtoenhancecollaborationbetweenbusinessandtechnologyteams.8
[1]
/The-Age-of-AI-Has-Begun
[2]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
[3]CapgeminiResearchInstitute“GenerativeAIinOrganizations”,July2023
[4]GlobalAISoftwareForecast,ForresterResearch,Inc.September29th,2022
[5]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
[6]Bytotalproductivityimprovementwemeanoverallimprovement
intheproductivityoftheindividualfromalltypesoftasksacceleratedbygenerativeAI.
[7]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
[8]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
4AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Bringingtomorrow’ssoftwareengineering
paradigmintoview:augmentedsoftwareteams
WebelievegenerativeAIwill
graduallytransformtheway
softwareisdeveloped.Theadventoflargelanguagemodels(LLMs)hasintroducedacompelling
rationaleforaparadigmshifttomoreAI-assisted(augmented)
softwareteams.TheintegrationofgenerativeAIintosoftware
engineeringoffersthepromiseofsignificantlyelevatedproductivityandenhancedquality.The
foundationalprinciplesand
methodsofAgileandDevOpsareretained,includingcollaboration,
adaptability,timetovalue,productcentricity,andcontinuousfeedbackloops.
ByworkingwithAIassistants
poweredbyLLMs,augmented
softwareteamscanautomate
mundanetasks,andpromote
morenuanced,datadriven
decisionmaking.Thisoptimizesthesoftwarelifecycleandhelpstoachievemoremilestonesalongtheway.
Wecallthisapproach
conversationalsoftware
engineering.Thisiswheresoftware
teamsprompttheAIassistant,
askingittogeneratesnippetsofcode,troubleshootissues,orevenhelpindesigningandarchitectingthesoftware.Theideaisto
makethesoftwaredevelopmentworkflowmoredynamicand
interactive,throughcontinuous
conversationswiththeAIassistantinthedevelopmentenvironment.It’sawaytostreamlinetasksandsolveproblemsmoreefficiently,
enablingsoftwareteamstofocusonmorecomplexandcreativeaspectsthatboostoverallproductivity.
Themodernevolutionofpair
programmingwithgenerative
AIAugmentedsoftwareteams
caneffectivelybenefitfromand
optimizeusageofgenerativeAI
througharevisedorganizationandnewwaysofworking,guidedby
thefollowingprinciples:
•Augmentedpairprogramming:
Thebasicunitofanaugmentedsoftwareteamisoneormultiplepairsofsoftwareengineers
workingintermittentlywitheachotherorindividuallywiththe
AIassistant.
•AI-humancollaboration:TeammembersworkindividuallywiththeAIassistanttoautomate
repetitivetasks,understandandsolveproblems,orbrainstorm
ideas,leveraginganLLM’sspeedandknowledge,whileensuringhumancreativity.
•Human-humancollaboration:
AfterconversationswithanAIassistant,humansreviewthegeneratedoutputandenhanceitsquality.
•Seniorcoordination:Aseniorleadoverseestheaugmentedteam,coordinatingefforts
andresolvingconflicts,while
ensuringcontrolandvalidationtopromoteasmoothworkflowandhealthyteamdynamic.
[9]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
AccordingtoCapgeminiResearchInstitute,thebiggestgapin
essentialprerequisitesisusually
accesstoplatformsandtools,
includingIDEs,automationand
testingtools,andcollaboration
tools.Only27%oforganizationsclaimtohaveaboveaverage
availabilityofthese.Oftesting
domainprofessionals,24%say
theyhaveaccesstothesetools,
comparedto19%ofproject
andprogrammanagement
professionals,furtherhighlightinggapswithinthesoftware
engineeringfunction.9
5AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
DeepdiveonCodingAssistants
Applyingthetechnologytosoftwareengineeringwillsignificantlyassistsoftwareteamsinthemultitudeoftaskstheyperformacrossthetraditionalsoftwaredevelopmentlifecycle(SDLC).
Forexample,softwareengineersusegenerativeAIwhentransformingbacklogstoriesintosoftware,throughdesignandcodingactivities.GenerativeAIcanbeharnessedtocreatedesignoutputssuchasuserinterfacemockups,
entitymodels,andapplicationprogramminginterfaces(APIs).Thisleadstoasignificantproductivityimprovementwithoutcompromisingquality,asdesignoutputsarealwaysreviewed,updated,andvalidatedbysoftwareengineersortechnologyleads.
AlthoughadoptionofgenerativeAIforsoftwareengineeringisstillinitsearlystages,functions.GenerativeAIisexpectedtoplayakeyroleinaugmentingsoftwareworkforcewithbetterexperience,toolsandplatforms,andgovernance,assistinginmorethan25%ofsoftwaredesign,development,andtestingworkby2026.10
GenerativeAIisalsopoisedtoredefineconventionalprogrammingpracticesbyshiftingthefocusfromcodingtopromptengineeringandcodeproofreading,asconfirmedbyAndrejKarpathy,anOpenAIcomputerscientist,whorecentlysaid:“thehottestnewprogramminglanguageisEnglish.”11Asanexample,usingplainlanguage,softwareengineerscandescribetheintendedfunctionalityofasoftwarefeature,thenreview,update,andvalidatethe
generatedoutput.Therearemanyotherexamples,suchasauto-completionofcode,generatingcodeforunittesting,(retro)documentation,andcodemigrationfromonelanguagetoanother.
AveryimportantvalueofgenerativeAIisthatitsupportsdevelopersalreadyduringcoding.Itcaneithersuggestcleancodedirectlyorevaluateexistingcodetoimprovesoftwarequalityifitfindsissues.
[10]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
[11]
/karpathy/status/1617979122625712128?lang=en-GB
6AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Howtomoveforward:
Aprovenmethodforasoftwareengineeringtransformation
Tostartthisjourneywith
generativeAI,organizations
willreaphugebenefitsfrom
partneringwithanexperiencedandtrustedadvisor.InadditiontohavingconsiderableAIexpertise,thispartnershouldbereadyto
overseeexperimentationsandformcollaborativeprojectteamsthateventuallyflourishintojointsoftwarehouses.
Avarietyoffactors,suchas
companysize,staffingpyramid,industry-specificgoals,meanthateachorganizationalunit
progressesatadifferentpaceandwithadifferentmodel.Therefore,eachwillhavedistinctneedsandchallengestoconsiderbefore
generativeAIcanbecomeapartofdailyoperations.
It’simportanttofocusonthreeareas:
•Assesstheorganization’smaturityanddeveloparoadmap.
•Runexperimentationsandmeasureimpact.
•Deploymentatscale.
Assesstheorganization’smaturityanddeveloparoadmapwithclearobjectives.
Withinacompany,thereare
usuallymanyvariationsofanSDLC.Typically,everyprogram/project/producthasitsownSDLCversion.Ameticulousassessmentand
deepunderstandingoftheSLDCofanITdomainwillrevealthe
currentmaturityofitssoftwareengineeringprocesses.And
whethertheyalignwithindustrybestpractices.Theevaluation
willalsoidentifyareasfor
improvement,particularlywherebottlenecksorinefficienciesoccur.
Basedonthisassessment,
objectivescanbedefinedby
selectingthemostpromising
transformationenablers,aswellasidentifyingtheirassociatedrisks
andchallenges.SomeenablerswillbepoweredbygenerativeAIwhile,forothers,itwillbemoreamatterofapplyingsoftware
engineeringbestpractices.It’simportanttodefinekeymetrics
tomeasuretheoutcomesofthefuturetransformationandinvolvestakeholdersearlytofostera
collaborativeenvironment.
Finally,acomprehensivevalue,
accessibility,andriskanalysiswillserveasabasisforestablishingaroadmapforchange.Alloptionswillbeweightedandprioritized
tomakeinformeddecisionsaboutresourceallocationasdifferent
domains,teams,androlesaredefinedforeachoption.
7AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Runreal-worldexperimentationsandmeasure
theimpactofGenerativeAI
Withtheprioritiesset,it’stime
toselectthebestgenerativeAI
toolstoachievethem.Andput-upguardrailstomanageanylegal
andcybersecurityrisks,while
controllingcosts.Measurementandcontinuousimprovementwillbepivotal.Asgenerative
AIisinsertedintomoreSDLC
processes,organizationsneed
togaugeitsimpact.Thismeansmeasuringcriticalaspectssuch
asproductivityenhancements,
softwarequality,time-to-market,anddeveloperexperience.A
feedbackloopshouldbeputinplacesothepaceandscopeofthedeploymentcanbeadaptedtoaccountforinefficiencies.Orrespondtoevolvingneedsandspecificobjectives.Critically,
theexperimentationsand
measurementsshouldberelevanttothebusinesscontextand
industryenvironmentoftheorganization.
Buttobetrulycompetitive,
organizationsneedmorethan
internalmeasurement.At
Capgemini,wehavedevelopedanindustrializedvaluemeasurementprotocolthatevaluatesthe
objectiveimpactofgenerative
AIacrossanorganization’smanySDLCs.Itisusedtomeasureandcompareanorganization’smetricsagainstourbenchmark,which
factorsinallourrelatedinternal
andexternalgenerativeAIprojects.Thisgivesorganizationsaclearviewofhowtheystackupagainsttheirpeers.
Toputthisintoperspective,
theadoptionofgenerativeAI
forsoftwareengineeringisstill
initsearlystageswith9in10
organizationsyettoscale.27%
oforganizationsarerunning
generativeAIpilots,and11%havestartedleveraginggenerativeAIintheirsoftwarefunctions.
Threeinfour(75%)large
organizations(annualrevenuegreaterthan$20billion)haveadopted(piloted/scaled)
generativeAIcomparedto23%oftheirsmallercounterparts(annualrevenuebetween$1–5billion).12
DeployGenerativeAIforSoftwareEngineering
atscale
Afterexperimentingwith
generativeAIthroughreal-world
pilotsandfullscaledelivery,
manyorganizationswillwantto
broadentheirapplicationsand
possiblyinvolvehundredsoreventhousandsofdevelopers.For
suchlargedeployments,careful
considerationmustbegiventotheorganizationalandHRimplications.
Sincevariousrolesmayneedto
change,it’simportanttointroduceagenerativeAIupskillingprogramthatwillhelpshapethenew
softwareengineeringpyramid,
andaddressthewaysofworkinginrespecttotheskillsrequired.Thiswilllaythefoundationfor
astrategicplanthatgradually
integratesgenerativeAIintothesoftwarelifecycle.Afurtherstep
maybetooffercoachingand
assistancetosteeremployeesastheypreparetoworkalongsideahostofnewgenerativeAItools.
Finally,adedicatedteamshould
besetuptodefinebusinesscases,measureprogress,andensuretheoutcomesmeetexpectations.Thisisessential,asthepaceandscopeofeachdeploymentwillinvariablyneedadjustment.
[12]CapgeminiResearchInstitute“Turbochargingsoftware”,June2024
8AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Capgemini’sexperienceinmeasuring
GenerativeAIimpact
Ourexperienceworkingwith
clientsacrossallsectorstells
usthatit’snotaquestionofif
generativeAI-poweredsoftwareengineeringwilldisruptand
reinventanorganization,but
ratherwhen,howquickly,andhowradically.Therearetwoquestionswehearthemostfromour
clients.First,howwillgenerative
AI-basedassistanceaffectsoftwareproductivityandquality.And
second,howwillsoftwareteams’organizationandwaysofworkingchangebecauseofit?
Atthebeginningof2023,
Capgeministartedalarge-scaleglobalprogramtoexperimentwithusecasestomeasure
generativeAI’simpactthrough
experimentations,bothinternallyandjointlywithclients.Theaimwastoseewhereandhowgenerative
AIcanaugmentthemanytasks
softwareteamstypicallyperform.Wedeployedareliableand
consistentmeasurementprotocolandnowweareprogressively
consolidatingourresultsintoa
repositoryforinternalandexternalusage/benchmarking.
Measurementprotocolandreal-worldexperimentations:designedforactionableand
reliableresults
Measuringproductivitywithin
softwaredevelopmentposes
inherentcomplexitiesdueto
themultifacetednatureoftheSDLC,thedynamicenvironmentitoperatesin,andthesubjective
andintangibleaspectsofmanyofitscomponents.Acomprehensivemeasurementapproachmust
encompassbothqualitativeand
quantitativefactors,tailoredtothespecificcontextoftheproject.
9AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Similarly,assessingsoftware
qualitypresentschallengesasit
involvesvariousdimensionssuchasfunctionality,performance,
reliability,usability,maintainability,security,andscalability,each
requiringitsownsetofmetricsandcriteria.
Moreover,solicitingfeedback
fromsoftwareengineersutilizingGenerativeAIonadailybasisis
crucial,consideringitsimpacton
thedevelopmentenvironmentandwoΓkHow.TheΓefoΓe,it’simpeΓativetodeviseandimplementapracticalmeasurementprotocol.Thisshouldbefocusedonunderstanding
Ge∩eΓativeAl’si∩Hue∩ceo∩codi∩9andunittestingwithinbespoke
applicationdevelopment,providingacleaΓi∩si9hti∩toitse阡ects.
Themeasurementprotocolisa
combi∩atio∩ofseveΓaldi阡eΓe∩ti∩9Γedie∩tsa∩dawell-defi∩ed
processtoproducetocreate
comprehensible,actionableandreliableresults.Thisincludesa
measurementapproach,metrics,team,a∩dawell-defi∩edpΓocess.
Measurementapproach:A
robustmeasurementapproach
includesstepsformeasuring
progress,includingplanning,
settingbaselines,andrunning
anexperimentation.Theright
measurementmetricsarekey,
andwellexploretheminthenextchapter.Tosupporttheapproach,metricstoolslikeSonarQube,
CAST,Jira,anddevelopersurveysareusedforcollectingand
analyzingdata.Prerequisites
andsuccessfactorsforaproperandconsistentmeasurement
areteamstability,duration,
technology,legalconsiderations,
andcybersecurity.Anormalization
processmusthandlevariability
duringexperimentationexecution,adjustingmetricstochangesin
teamsize,capacity,orcomplexity.
Metrics:Codingvelocityservesasapivotalmetrictomeasureteamproductivity,focusingoncoding
andunittestingactivities,typicallyqua∩tifiedbyimpleme∩tedstoΓypoints.Otherdimensionsmust
alsobeevaluated-likevelocityperdeveloperseniorityandvelocity
percomplexityofuserstories.ThiscompΓehe∩siveappΓoacho阡eΓs
insightsintohowGenerativeAIi∩Hue∩cespΓoductivityacΓoss
di阡eΓe∩tdevelopeΓskilllevelsandtheintricaciesofsoftwaredevelopmenttasks.
Besidesthat,whenitcomesto
testing,themetricforUnitTest
Coverageisessential.Ithelpstoassessthequalityandreliability
ofsoftware.Tokeepitsimplewefocusoninstructioncoverage(C0)asthisismeasuredbymostofthetools.
Ontopofthoseourprotocol
includesmanymoremetricslike
codee代cie∩cy,codesecuΓity,codesmells,codeduplication.
Team:Inthesingleteamapproach,oneteamsequentiallyexecutes
abacklogofuserstoriesof
consistentsizeorcomplexity,
comparingperformancewithandwithoutGenerativeAIassistance.
Ontheotherhand,themultiple
teamsapproachinvolvesparallel
executionofthesamebacklogbyatleasttwoteamswithdi阡eΓe∩t
toolsetups.Forexample,onewithGenerativeAItoolsandtheother
withoutGenerativeAItools.This
allowsforsimultaneousassessmentofGe∩eΓativeAl’se阡ectsacΓoss
di阡eΓe∩tteamdy∩amics.
Theseniorityorcapabilities
ofateamareimportantfor
normalization,andsoits
mandatorytoknowwhatkindofteammixisworkingonthedefi∩edbacklo9.Wedisti∩9uishaseniorteamofhighlyskilledindividualsrepresentingthe
9oldsta∩daΓdofpΓoficie∩cy.
Conversely,thewell-balanced
teamconsistsofagoodmixof
seniorsandjuniors,necessitatingsomelevelofcoachingtypically
facilitatedbyseniormembers
alongsidedailywork.Finally,the
juniorteamfeaturesfewseniors,withtheprimaryfocusoncoachingjuniormembersdevelopment
andproductivity.
Process:Oncealltheingredientshavebee∩defi∩ed,apΓocessisneededtoensurehighquality
Γesultsa∩dΓeducesidee阡ectsduetoestimationinaccuracyandthe
humanfactor.
1.Defi∩etheteam(s)oΓ9a∩izatio∩andtheexperimentationscopeandtimeline.
2.Defi∩ethemeasuΓeme∩tapproach.
3.Validatetheprerequisitesandthesuccessfactors.
4.Conductabaselineforthemetrics,without
GenerativeAIassistance.
5.Executetheexperimentationsprintswith
GenerativeAIassistance.
6.Collectandnormalizethemetricsandthefeedback.
7.Consolidateandreportthemeasuredresults.
10AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Prerequisitesandsuccessfactors
Successfulexperimentation
hingesonmanagingthefollowingprerequisiteseffectively:
•Teamstabilityisvitalfor
achievingcomparableresults.Thisincludesconsistentsize,seniority,andunwavering
processes.
•Baseliningnecessitatesa
minimumofthreesprints,whileGenerativeAIexperimentationsbenefitfrom6-9sprints.Awell-definedbacklogwithdiverse
userstoriesisessentialinputforthesprintsandshouldbeoperatedaccordingly.
•Measurementtoolsmustbe
readilyavailableforanaccurateassessment.Allteammembersneedaccess.Compliancewith
regulations,encompassinglegal,security,andprivacyaspects,
isimperativethroughouttheprocess.
Baseliningiscrucialfor
understandingthecurrentvelocityandqualityofthetraditionalwaysofworkingwithoutGenerative
AItools.It’srecommendedthat
insightsaregatheredfromthe
lastthreesprints,excluding
GenerativeAIassistance.Coding
velocitydatashouldbegatheredfromteambacklogmanagementandcollaborationtools,aswell
ascodequalityreportsfrom
staticcodeanalysistools.Ifdata
fromthelastthreesprintsisn’t
available,threesprintswithout
GenerativeAIassistancemustbeconductedbeforeproceeding
totheGenerativeAIphase.The
baselinemustbecalculatedbasedonthis.Additionally,thedeveloperexperiencesurvey,basedonthe
experiencewithoutGenerativeAI,isconductedduringthe
baselinephase.
Forasuccessfulexperimentation,it’simportanttomaintaina
constantteamsizeandcarefully
curatethebacklogtoincludeusecasesofvaryingcomplexities.Theteammustbeequippedwiththenecessarysetuptoeffectively
utilizetheGenerativeAItool.
Additionally,assigningadedicatedtoolexperttoeachteamcan
significantlyenhancesuccessandefficiency.Monitoringtheteam
involvesensuringtimelytask
updatesonplatformslikeJIRAorAzure,withteamleadsresponsiblefortrackingandensuring
compliance.Inanaugmented
teamapproach,peerreviewsareessential,andtheGenerative
AItoolshouldbeutilizedasan
assistantthroughouttheprocess.
11AugmentedSoftwareTeams:HowtodrivemaximumassistancefromGenerativeAI
Lessonslearned
Likeallorganizations,weare
learningfromGenerativeAIinrealworldandrealsizeconditions.
Belowareourhigh-levellearnings:
ExperimentationFramework:OurexperimentationswithgenerativeAIincustomsoftwareengineeringtypicallyinvolveonetomultiple
teams,spanningaminimum
durationofsixsprints(preferablynine),withadiversebacklogof
userstoriesandcomplexitymix.
Differentteamconfigurations
offervaluableinsights.Typical
configurationsmightinclude,
existingteamstransitioningto
usinggenerativeAI.Newteams
integratinggenerativeAIlater.Andshadowteamsworkingalongsideexistingones.
BaselineEstablishment:Solid
baseliningiscrucialtoensure
robustandrepresentative
comparisonsacrossvarious
metrics.Theseincludevelocity,quality,security,time-to-market,anddeveloperexperience.
Quantifiableresultsfrom
experimentationshowcaseproductivityimprovements.
ToolingPerspective:
Experimentationsinthecodingandunittestingstagesofthesoftwarelifecyclecovermultipletechnologycombinations.Thisincludesvendor-packagedsolutionswithgenerativeAIextensionsandfoundational
LLMsenhancedwithcontextualizedpromptengineeringtechniques.
Pro
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 熔炉自动化控制系统的维护考核试卷
- 畜牧业智能监控系统设计与实施考核试卷
- 山西省晋城市高平市2025年数学五下期末学业水平测试试题含答案
- 江苏省无锡市长泾片2024-2025学年初三毕业班“三诊”模拟考试英语试题试卷含答案
- 绵阳市2025届四年级数学第二学期期末达标检测模拟试题含解析
- 四川省自贡市曙光中学2025届初三年级模拟考试(二)数学试题含解析
- 南昌工学院《文学创作与评论训练》2023-2024学年第二学期期末试卷
- 南京邮电大学通达学院《油气储运新技术》2023-2024学年第二学期期末试卷
- 武汉海事职业学院《园林专业英语》2023-2024学年第一学期期末试卷
- 沈阳城市学院《轻松学营销》2023-2024学年第二学期期末试卷
- 2024年学校家校关系纠纷应急处置预案
- 机房建设工程方案
- 初中学习经验分享
- 麦肯锡的《问题分析与解决技巧》课件
- 职业教育技能培训项目化
- 西装基础知识课件
- 提高术前准备完善率品管圈课件
- 中国自闭症儿童现状分析报告
- 家禽屠宰厂的可行性方案
- 情绪心理学与情绪管理 课件
- 汽车故障诊断技术第3版微课张钱斌课后参考答案
评论
0/150
提交评论