Ukukala imisebenzi yokubuyekezwa kokuqukethwe ngokugeleza komsebenzi kwama-agent amaningi

Amabhizinisi aphethe amavolumu akhula njalo wokuqukethwe, kusukela kumakhathalogi emikhiqizo nama-athikili asekelayo kuya kuzisekelo zolwazi kanye nemibhalo yezobuchwepheshe. Ukuqinisekisa ukuthi lolu lwazi luhlala lunembile, lufanelekile, futhi luhambisana namaqiniso akamuva ebhizinisi kuyinselele enkulu. Izinqubo zokubuyekezwa kokuqukethwe mathupha zivame ukuhamba kancane, zibiza, futhi azikwazi ukuhambisana nezidingo zebhizinisi eziguqukayo. Ngokocwaningo lukaMcKinsey, izinhlangano ezisebenzisa i-AI ekhiqizayo ngomsebenzi wolwazi, okuhlanganisa ukubuyekezwa kokuqukethwe nokuqinisekiswa kwekhwalithi zingakhuphula umkhiqizo ngokufika ku-30–50% futhi zinciphise ngokumangalisayo isikhathi esichithwa emisebenzini yokuqinisekisa ephindaphindwayo. Ngokufanayo, ucwaningo oluvela kwa-Deloitte lugqamisa ukuthi ukusebenza kokuqukethwe okushayelwa yi-AI akugcini nje ngokukhuphula ukusebenza kahle kodwa futhi kusiza izinhlangano zigcine ukunemba okuphezulu kokuqukethwe futhi kunciphise ubungozi bokusebenza.

I-Amazon Bedrock AgentCore, ingqalasizinda eyakhelwe inhloso yokuphakela nokusebenzisa ama-ejenti e-AI esikalini, ehlanganiswe nama-Strands Agents, i-SDK yomthombo ovulekile wokwakha ama-ejenti e-AI, inika amandla izinhlangano ukuthi zizenzele ngokuzenzakalelayo ukugeleza kokusebenza kokubuyekezwa kokuqukethwe okuphelele. Le ndlela esekelwe kumenzeli inika amandla amabhizinisi ukuthi ahlole okuqukethwe ngokunemba, aqinisekise ulwazi ngokumelene nemithombo egunyaziwe, futhi enze izincomo ezingenziwa ukuze kuthuthukiswe. Ngokusebenzisa ama-ejenti akhethekile asebenza ndawonye ngokuzenzakalelayo, ochwepheshe babantu bangagxila emisebenzini yokubuyekezwa yamasu kuyilapho isistimu ye-ejenti ye-AI iphatha ukuqinisekiswa kokuqukethwe okukhulu.

Indlela esekelwe kumenzeli esiyethulayo isebenza kunoma yiluphi uhlobo lokuqukethwe kwebhizinisi, kusukela ekubhalweni komkhiqizo nezisekelo zolwazi kuye ezintweni zokumaketha kanye nokucaciswa kobuchwepheshe. Ukuze sibonise le miqondo isebenza, sidlula esibonelweni esisebenzayo sokubuyekeza okuqukethwe kwebhulogi ukuze sithole ukunemba kobuchwepheshe. Lawa maphethini namasu angashintshwa ngokuqondile ezidingweni ezihlukahlukene zokubuyekezwa kokuqukethwe ngokulungisa ukulungiselelwa kwe-ejenti, amathuluzi, nemithombo yokuqinisekisa.

Uhlolojikelele lwesixazululo

Isixazululo sokubuyekezwa kokuqukethwe sisebenzisa iphethini yokuhamba komsebenzi enabenzeli abaningi, lapho ama-ejenti amathathu akhethekile e-AI akhelwe ngama-Strands Agents futhi asetshenziswa ku-Amazon Bedrock AgentCore esebenza ngepayipi elihlanganisiwe. I-ejenti ngayinye ithola okukhiphayo kumenzeli wangaphambili, ikucubungule ngokomsebenzi wayo okhethekile, bese idlulisela ulwazi olunothisiwe kumenzeli olandelayo ngokulandelana. Lokhu kudala inqubo yokuhluza eqhubekayo lapho:

1. I-ejenti yesithwebuli sokuqukethwe ihlaziya okuqukethwe okungahluziwe futhi ikhiphe ulwazi olufanele
2. Umenzeli wokuqinisekisa okuqukethwe ithatha lezi zakhi ezikhishiwe futhi iziqinisekise ngokumelene nemithombo egunyaziwe
3. I-ejenti yokuncoma iguqula okutholiwe kokuqinisekisa kube izibuyekezo zokuqukethwe okusetshenzwayo

Ukugcinwa kokuqukethwe kobuchwepheshe kudinga ama-ejenti akhethekile ngenxa yokuthi ukuskena mathupha, ukuqinisekisa, nokubuyekeza amadokhumenti akuphumelelanga futhi kuba namaphutha. I-ejenti ngayinye inendima egxilile – isithwebuli sihlonza izici ezizwela isikhathi, isiqinisekisi sihlola ukunemba kwamanje, futhi i-ejenti yesincomo yenza izibuyekezo ezinembile. Idizayini ye-modular yohlelo, enokuxhumana ecacile nezibopho, yenza kube lula ukwengeza ama-ejenti amasha noma ukwandisa amakhono njengoba ubunzima bokuqukethwe bukhula. Ukubonisa ukuthi lolu hlelo lokubuyekezwa kokuqukethwe okusekelwe kumenzeli lusebenza kanjani, sidlula ekusetshenzisweni okubuyekeza okuthunyelwe kwebhulogi lobuchwepheshe ngokunemba. Izinkampani zobuchwepheshe zivame ukushicilela okuthunyelwe kwebhulogi okuchaza izici ezintsha, izibuyekezo, nemikhuba engcono kakhulu. Kodwa-ke, isivinini esisheshayo sokusungula izinto sisho ukuthi ezinye izici ziyahoxiswa noma zibuyekezwe, okwenza kube inselele ukugcina ulwazi lukhona kumakhulu noma izinkulungwane zokuthunyelwe okushicilelwe. Nakuba sibonisa le phethini ngokuqukethwe kwebhulogi, i-architecture ayinalwazi lokuqukethwe futhi isekela noma yiluphi uhlobo lokuqukethwe ngokulungiselela ama-ejenti ngemiyalelo efanele, amathuluzi, nemithombo yedatha.

Isibonelo esisebenzayo: Isixazululo sokubuyekeza okuqukethwe kwebhulogi

Sisebenzisa ama-ejenti amathathu akhethekile axhumana ngokulandelana ukuze abuyekeze ngokuzenzakalelayo okuthunyelwe futhi ahlonze ulwazi lobuchwepheshe oluphelelwe yisikhathi. Abasebenzisi bangacupha isistimu ngokwenza noma bayihlele ukuthi isebenze ngezikhathi ezithile.

Isakhiwo sokubuyekeza okuqukethwe kwebhulogi Figure-1

Ukugeleza komsebenzi kuqala lapho i-URL yebhulogi inikezwa umenzeli wesithwebuli sebhulogi, othola okuqukethwe kusetshenziswa ama-Strands http_request ithuluzi futhi ikhiphe izimangalo zobuchwepheshe ezibalulekile ezidinga ukuqinisekiswa. Umenzeli wokuqinisekisa ube esebuza iseva ye-MCP yemibhalo ye-AWS ukuze ilande amadokhumenti akamuva futhi iqinisekise izimangalo zobuchwepheshe ngokumelene nemibhalo yamanje. Okokugcina, i-ejenti yesincomo ihlanganisa okutholakele futhi ikhiqize umbiko ophelele wokubuyekeza onezincomo ezisebenzayo zethimba lebhulogi.

Ikhodi ingumthombo ovulekile futhi isingathwe ku-GitHub.

Ukuhamba komsebenzi kwama-agent amaningi

I-ejenti yesithwebuli sokuqukethwe: Ukukhishwa okuhlakaniphile kokutholwa kokuphelelwa yisikhathi

I-ejenti yesithwebuli sokuqukethwe isebenza njengendawo yokungena ekuhambeni komsebenzi wama-agent amaningi. Inesibopho sokuhlonza ulwazi lobuchwepheshe okungenzeka luphelelwe yisikhathi. Lo menzeli uqondise ngokuqondile izici okungenzeka ziphelelwe yisikhathi ngokuhamba kwesikhathi. I-ejenti ihlaziya okuqukethwe futhi ikhiqize okukhiphayo okuhlelekile okuhlukanisa isici ngasinye sobuchwepheshe ngohlobo, indawo kubhulogi, kanye nokuzwela kwesikhathi. Le fomethi ehlelekile yenza i-ejenti yokuqinisekisa ukuthi yamukele idatha ehlelwe kahle engayicubungula kahle.

I-ejenti yokuqinisekisa okuqukethwe: Ukuqinisekisa okusekelwe ebufakazini

I-ejenti yokuqinisekisa okuqukethwe ithola izinto zobuchwepheshe ezihleliwe kumenzeli weskena futhi yenza ukuqinisekiswa ngokumelene nemithombo egunyaziwe. I-ejenti yokuqinisekisa isebenzisa iseva ye-MCP yemibhalo ye-AWS ukuze ifinyelele imibhalo yezobuchwepheshe yamanje. Ngento ngayinye yobuchwepheshe etholwe kumenzeli weskena, ilandela inqubo yokuqinisekisa ehlelekile eholwa ukwaziswa okuthile okugxile kunhloso, imibandela elinganisekayo.

I-ejenti iyalwa ukuthi ihlole:

• Ulwazi oluqondene nenguqulo: Ingabe inombolo yenguqulo eshiwo, indawo yokugcina ye-API, noma ipharamitha yokumisa isekhona?
• Ukutholakala kwesici: Ingabe isici sesevisi esichazwe sisatholakala ezifundeni ezishiwo noma izigaba?
• Ukunemba kwe-syntax: Ingabe izibonelo zekhodi, imiyalo ye-CLI, noma amazwibela okumisa afana nemibhalo yamanje?
• Ukufaneleka okufunekayo: Ingabe izidingo ezisohlwini, okuncikile, noma izinyathelo zokusetha zisanembile?
• Amanani nemikhawulo: Ingabe izindleko ezishiwo, izabelo, noma imikhawulo yesevisi ihambisana nolwazi olushicilelwe manje?

Ngento ngayinye yobuchwepheshe etholwe kumenzeli weskena, umenzeli wenza lezi zinyathelo ezilandelayo:

1. Ikhiqiza imibuzo yosesho eqondiwe ngokusekelwe ohlotsheni lwe-elementi nokuqukethwe
2. Ibuza iseva yemibhalo ngolwazi lwamanje
3. Iqhathanisa isimangalo sokuqala esibhekene nemithombo egunyaziwe kusetshenziswa le mibandela engenhla
4. Ihlela umphumela wokuqinisekisa ngokuthi CURRENT, PARTIALLY_OBSOLETEnoma FULLY_OBSOLETE
5. Ibhala ukungqubuzana okuthile nobufakazi

Isibonelo sokuqinisekisa siyasebenza: Lapho umenzeli weskena ehlonza isimangalo esithi “I-Amazon Bedrock itholakala ezifundeni zase-us-east-1 kanye nase-us-west-2 kuphela,” Umenzeli Wokuqinisekisa ukhiqiza umbuzo wosesho “izifunda ezitholakalayo ze-Amazon Bedrock” futhi ithola ukutholakala kwamanje kwesifunda kumadokhumenti e-AWS. Ngemva kokuthola ukuthi i-Bedrock isiyatholakala ezifundeni ezingu-8+ ezihlanganisa i-eu-west-1 kanye ne-ap-southeast-1, ibeka lokhu kusigaba ngokuthi PARTIALLY_OBSOLETE enobufakazi: “Isimangalo sangempela sibonisa izifunda ezingu-2, kodwa imibhalo yamanje ibonisa ukutholakala e-us-east-1, us-west-2, eu-west-1, ap-southeast-1, kanye nezinye izifunda ezingu-4 kusukela ngosuku lokuqinisekisa.”

Okukhiphayo komenzeli wokuqinisekisa kugcina isakhiwo sesici esivela kumenzeli weskena kuyilapho yengeza le mininingwane yokuqinisekisa kanye nezigaba ezisekelwe ebufakazini.

I-ejenti yezincomo: Ukwenziwa kwesibuyekezo esingasetshenzwa

I-ejenti yesincomo imelela isigaba sokugcina sokugeleza komsebenzi kwama-agent amaningi, eguqula okutholiwe kokuqinisekisa kube izibuyekezo zokuqukethwe osekulungele ukuqaliswa. Lo menzeli uthola imiphumela yokuqinisekisa futhi ukhiqize izincomo ezithile ezigcina isitayela sokuqukethwe kwasekuqaleni kuyilapho elungisa amaphutha ezobuchwepheshe.

Ukulungisa iphethini yokuhamba komsebenzi enabenzeli abaningi ezimweni zakho zokusebenzisa ukubuyekezwa kokuqukethwe

Iphethini yokuhamba komsebenzi enabenzeli abaningi ingashintshwa ngokushesha kunoma yisiphi isimo sokubuyekezwa kokuqukethwe ngaphandle kwezinguquko zezakhiwo. Ukuthi ubuyekeza amadokhumenti omkhiqizo, izinto zokumaketha, noma imibhalo yokuthobela imithetho, kuyasebenza ukuhamba komsebenzi okulandelanayo kwama-ejenti amathathu. Ukwaziswa kwesistimu kudinga ukushintshwa ukuze umenzeli ngamunye agxile ezintweni ezithile zesizinda futhi akwazi ukushintsha amathuluzi noma imithombo yolwazi. Isibonelo, ngenkathi isibonelo sethu sokubuyekezwa kwebhulogi sisebenzisa i http_request ithuluzi lokulanda okuqukethwe kwebhulogi kanye Neseva ye-AWS Documentation MCP ukuze kuqinisekiswe, isistimu yokubuyekeza ikhathalogi yomkhiqizo ingase isebenzise ithuluzi lesixhumi sesizindalwazi ukuze ithole ulwazi lomkhiqizo kanye nama-API okuphatha uhlu lwemibuzo ukuze kuqinisekiswe. Ngokufanayo, isistimu yokubuyekeza ukuthobelana ingalungisa ukwaziswa komenzeli weskena ukuze akhombe izitatimende zokulawula esikhundleni sezimangalo zobuchwepheshe, ixhume i-ejenti yokuqinisekisa kusizindalwazi somthetho esikhundleni semibhalo yobuchwepheshe, futhi ilungise i-ejenti yokuncoma ukuthi ikhiqize imibiko elungele ukuhlolwa esikhundleni sezibuyekezo zokuqukethwe. Izinyathelo ezilandelanayo eziyinhloko zokukhipha, ukuqinisekiswa, nokuncoma kuhlala kungashintshile kuzo zonke lezi zimo, kuhlinzeka ngephethini eqinisekisiwe esukela kumabhulogi wezobuchwepheshe iye kunoma yiluphi uhlobo lokuqukethwe kwebhizinisi. Sincoma izinguquko ezilandelayo ukuze wenze ngokwezifiso isixazululo kwezinye izinhlobo zokuqukethwe.

1. Faka esikhundleni amanani we CONTENT_SCANNER_PROMPT, CONTENT_VERIFICATION_PROMPTfuthi RECOMMENDATION_PROMPT okuguquguqukayo nemiyalelo yakho yokwaziswa yangokwezifiso:

  python
  CONTENT_SCANNER_PROMPT = """"""
  CONTENT_VERIFICATION_PROMPT = """"""
  RECOMMENDATION_PROMPT = """"""

2. Buyekeza iseva ye-MCP yemibhalo esemthethweni ye-ejenti yokuqinisekisa okuqukethwe:

  python
   product_db_mcp_client = MCPClient(
       lambda: stdio_client(StdioServerParameters(
           command="uvx", args=[""]
       ))
   )

3. Engeza amathuluzi okufinyelela okuqukethwe afanelekile njenge database_query_toolfuthi cms_api_tool okomenzeli weskena sokuqukethwe nini http_request ithuluzi alanele:

python
   scanner_agent = Agent(
       model="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
       system_prompt=CONTENT_SCANNER_PROMPT,
       tools=[database_query_tool, cms_api_tool]  # Replace http_request
   )

Lezi zinguquko ezihlosiwe zinika amandla iphethini yezakhiwo efanayo ukuthi isingathe noma iluphi uhlobo lokuqukethwe kuyilapho kugcinwa ukwakheka kokugeleza komsebenzi kwama-ejenti amathathu afakazelwe, okuqinisekisa ukwethembeka nokuvumelana kuzo zonke izizinda zokuqukethwe ngaphandle kokudinga izinguquko kumqondo we-orchestration oyinhloko.

Isiphetho nezinyathelo ezilandelayo

Kulokhu okuthunyelwe, sichaze ukuthi singaklama kanjani i-agent ye-AI enikwe amandla uhlelo lokubuyekeza okuqukethwe kusetshenziswa i-Amazon Bedrock AgentCore kanye nama-Strands Agents. Sibonise iphethini yokuhamba komsebenzi enabenzeli abaningi lapho abasebenzeli abakhethekile besebenza ndawonye ukuze baskene okuqukethwe, baqinisekise ukunemba kobuchwepheshe ngokumelene nemithombo egunyaziwe, futhi benze izincomo ezingenzeka. Ukwengeza, sixoxisane ngendlela yokujwayela le phethini yama-agent amaningi ezinhlotsheni zokuqukethwe ezihlukene ngokushintsha ukwaziswa komenzeli, amathuluzi, nemithombo yedatha kuyilapho sigcina uhlaka lwezakhiwo olufanayo.

Sikukhuthaza ukuthi uhlole isampula yekhodi etholakala ku-GitHub ku-akhawunti yakho ukuze uthole ulwazi lokuqala ngesixazululo. Njengezinyathelo ezilandelayo, cabanga ukuqala ngephrojekthi yokuhlola engxenyeni yokuqukethwe kwakho, ukwenza ngendlela oyifisayo ukwaziswa kwe-ejenti yesizinda sakho esithile, nokuhlanganisa imithombo yokuqinisekisa efanele yesimo sakho sokusebenzisa. Imvelo ye-modular yalesi sakhiwo ikuvumela ukuthi wenze ngcono ngokuphindaphindiwe amakhono e-ejenti ngayinye njengoba unweba isistimu ukuze isingathe izidingo zokubuyekeza okuqukethwe okugcwele kwenhlangano yakho.

Mayelana nababhali

USarath Krishnan uyi-Senior Gen AI/ML Specialist Solutions Architect e-Amazon Web Services, lapho esiza amakhasimende ebhizinisi ukuthi aklame futhi akhiphe i-AI ekhiqizayo nezixazululo zokufunda ngomshini eziletha imiphumela yebhizinisi elinganisekayo. Uletha ubuchwepheshe obujulile ku-Generative AI, Ukufunda Ngomshini, kanye nama-MLOps ukuze akhe amasistimu e-AI angakala, avikelekile, futhi alungele ukukhiqiza.

Santhosh Kuriakose ungumakhi we-AI/ML Specialist Solutions Architect e-Amazon Web Services, lapho esebenzisa khona ubuchwepheshe bakhe ku-AI ne-ML ukuze akhe izixazululo zobuchwepheshe eziletha imiphumela yebhizinisi yamasu kumakhasimende akhe.

Ravi Vijayan Ungumphathi Wezixazululo Zekhasimende nge-Amazon Web Services. Uletha ubuchwepheshe njengonjiniyela, uMphathi Wohlelo Lobuchwepheshe, kanye Nozakwethu Weklayenti, futhi okwamanje ugxile ekusizeni amakhasimende abone ngokugcwele amandla nezinzuzo zokuthuthela efwini nokwenza isimanjemanje nge-Generative AI.