Ama-AI Agents: Balungela Kanjani Ukuqhumisa Isendlalelo Senqubo Yebhizinisi

Lapho abantu becabanga nge-AI ekhiqizayo, abaningi bayibona isebenza njengengxenye “yohlelo lokusebenzelana” - i-ejenti yesevisi yamakhasimende, ithuluzi lokuphatha i-supply chain, noma indlela yokuxhumana ngobuhlakani nokusesha ama-PDF enhlangano neminye imininingwane yokuphathelene.

Lokho umbono onembile – onyakeni olandelayo noma emibili, izinhlelo zokusebenza ezidala okuqukethwe ngobuhlakani ukusebenzisa amamodeli wezilimi ezinkulu (LLMs) izohlala igxile ku-AI eyinhloko yamabhizinisi.

Kodwa cabanga ngalokhu: Iningi lamakhodi abhalwe ezinkampanini lihlala ngaphakathi kwezinqubo zebhizinisi - amasistimu afana nokuhlelwa kwempahla, ahlala phakathi kwesendlalelo sokubandakanya kanye nezinhlelo zamarekhodi eziqinile (idatha yenhlangano, njll.) I-GenAI izowuzuza kanjani lowosendlalelo? Inhlangano ingathuthukisa kanjani izinqubo zayo zebhizinisi ngalobu buchwepheshe obudlangile?

I-Agent AI iyimpendulo. Nakuba ama-agent e-AI akhelwe ukwenza imisebenzi ethile noma enze imisebenzi ethile, ngokuvamile ephindaphindayo (njengokubuyekeza ikhalenda lakho), ngokuvamile adinga okokufaka komuntu. I-Agentic AI imayelana nokuzimela (cabanga ngezimoto ezizishayelayo), isebenzisa uhlelo lwama-ejenti ukuze lihlale lizivumelanisa nezimo eziguqukayo futhi lidale ngokuzimela, likhiphe, futhi lenze imiphumela ibe ngcono.

Uma i-agent AI isetshenziswa ekugelezeni kwenqubo yebhizinisi, ingangena esikhundleni sezinqubo zebhizinisi ezintekenteke, ezimile ngamasistimu ashintshashintshayo, aqaphela umongo.

Ake sibheke ukuthi kungani ukuhlanganisa ama-agent e-AI ekwakhiweni kwebhizinisi kuphawula ukweqa okuguquguqukayo endleleni izinhlangano ezibhekana ngayo nezinqubo ze-automation kanye nezinqubo zebhizinisi, nokuthi hlobo luni lwenkundla oludingekayo ukuze kusekelwe lezi zinhlelo zokuzenzakalela.

Okwenziwa Ama-ejensi Manje

Uma unikeza umenzeli onomongo, umenzeli ube esesuthisa lowo mongo ku-LLM bese eyicela ukuthi iqedele futhi iphendule kuyo. Abenzeli be-AI bangasebenzisa amakhono ukuqedela imisebenzi egameni labasebenzisi. Lawa ma-ejenti e-AI angenza imisebenzi eminingana ebalulekile eholwa yimiyalelo nolwazi olususelwa kumongo:

Ukusetshenziswa kwamathuluzi - I-ejenti isebenzisa imisebenzi yangaphandle, ama-API, noma amathuluzi ukuze inwebe amandla ayo futhi yenze imisebenzi ethile. Lokhu kungase kuhlanganise ukubiza imisebenzi echazwe kusengaphambili noma ukuxhumana namasevisi angaphandle (njengokwenza izicelo zewebhu usebenzisa i-cURL noma ukufinyelela ama-API angu-RESTful) ukuze uthole umongo noma wenze izenzo ezingaphezu kwemisebenzi yayo engokwemvelo.

Ukwenziwa kwezinqumo — I-ejenti ihlola ulwazi olutholakalayo bese ikhetha isenzo esifaneleka kakhulu ukufeza izinhloso zayo. Lokhu kuhlanganisa ukuhlaziya umongo, ukukala imiphumela engenzeka, nokukhetha inkambo yesenzo ehambisana nezinjongo ezifiswayo.

Ukuhlela — I-ejenti yenza ukulandelana kwezenzo noma amasu okufeza umgomo othile.

Ukubonisana — I-ejenti ihlaziya umongo okhona, ifinyelela iziphetho, ibikezele imiphumela yezenzo, futhi yenza izinqumo ezinolwazi mayelana nezinyathelo ezifanele okufanele zithathwe ukuze kufinyelelwe umphumela oyifunayo.

Lezi zinhlobo zokugcina zemisebenzi - ukwenza izinqumo, ukuhlela, nokubonisana - ngokuvamile zihlanganisa ama-ejenti amaningi asebenza ndawonye afinyelele umgomo. Ama-ejenti angafuna ukwenza ngcono ikhodi ekhiqiziwe ukuze afaneleke, aphikisane ngokuthi isinqumo somenzeli sichemile, noma ahlele ukusetshenziswa kwamanye amakhono e-ejenti ukuze aqedele umsebenzi.

Ukuhlela inethiwekhi ye-Agentic

Amamodeli amanethiwekhi wamandla ama-ejenti empeleni ayimisebenzi engenasimo ethatha umongo njengokufakiwe futhi okukhipha impendulo, ngakho uhlobo oluthile lohlaka luyadingeka ukuze uwahlele. Ingxenye yaleyo orchestra ingaba ukuthuthukiswa okulula (imodeli icela ulwazi olwengeziwe). Lokhu kungase kuzwakale kufana ne isizukulwane sokutholwa-engeziwe (i-RAG) - futhi kufanele, ngoba i-RAG iwuhlobo olwenziwe lula lwe-ejenti yokwakha: Ihlinzeka imodeli ngethuluzi elilodwa elithola ulwazi olwengeziwe, ngokuvamile oluvela kusizindalwazi se-vector.

Kodwa izinhlaka zemodeli ye-multiagent zithatha lokhu phambili; bacela umthengisi ukuthola ulwazi olwengeziwe noma banikeze impendulo edizayinelwe ukuthi inikezwe enye i-ejenti ukuze ithuthukiswe.

I-ejenti eyodwa ingakwazi, ithi, ibhale iPython, kuyilapho enye ibuye ibuyekeze. Noma i-ejenti ingase iveze umgomo noma umqondo, bese umsebenzi we-ejenti wesibili kube ukukuhlukanisa lokho kube isethi yemisebenzi noma ubuyekeze umqondo ukuze uthole izinkinga i-ejenti yokuqala engazibuyekeza, bese ihlunga umqondo. Imiphumela yakho iqala ukubangcono futhi ibengcono.

Kulungile, kodwa Ngiyakha Kanjani I-Agent AI?

Esikhathini esizayo esiseduze, onjiniyela abaningi be-software bazoba ababhali bezinqubo ze-agent. Bazokwakha lezi zinqubo ngokuxuba nokuqhathanisa izingxenye: amamodeli, okokufaka komsebenzisi, imigomo - nezinsizakalo zebhizinisi ezibalulekile.

Isibonelo sesinye salezi zingxenye isitokwe ohlelweni lokuphathwa kwempahla. Kuthiwani uma uxhuma leyo sistimu kumenzeli ongasiza ekuthuthukiseni amaleveli e-inventory ngesikhathi seholide? Ngosizo lomunye umenzeli owenze ukuhlaziya kwezinga lempahla yomlando, ungaqinisekisa ukuthi kunohlu lwamagama olwanele ukuhlangabezana nesidingo sesizini kodwa ushiye i-inventory encane ngemva kokujaha kweholide.

Lokhu kungase kudumaze abathengi abashisekayo bangemva kukaKhisimusi, kodwa kungasiza futhi ukuvimbela abathengisi ukuthi bathengise izinto zabo ngokulahlekelwa.

Kodwa onjiniyela bazozakha kanjani lezi zinhlelo?

Izinqubo ze-agent, vele, zingavezwa ngekhodi, kodwa futhi kusiza ukuzibona ngeso lengqondo “njengokugeleza kwe-ejenti” – okukhiphayo kwe-ejenti eyodwa kuba okokufaka kwenye i-ejenti, njalo njalo. Amathuluzi atholakalayo manje asevele enikeza inani elikhulu kulo mzamo wokwenza lula amasistimu wokwakha ama-ejenti. Isixazululo esisodwa esinjalo I-Langflow , umakhi obonakalayo, onekhodi ephansi yokwakha izinhlelo zokusebenza ze-AI ze-agent kanye nokugeleza komsebenzi we-AI okuyinkimbinkimbi ngokuhudula nokulahla izingxenye ezahlukene, ngaphandle kwesidingo sokubhala amakhodi okuningi.

I-Langflow inika amandla onjiniyela ukuthi bachaze noma yini njengethuluzi, okuhlanganisa izingxenye ezifana nokwaziswa, umthombo wedatha, imodeli, ama-API, amathuluzi, nanoma yimaphi amanye ama-ejenti. Sisanda kubona isidingo esibalulekile sokwakha “ukugeleza” ngama-ejenti, njengoba abathuthukisi bakha izinhlelo zokusebenza eziningi ezihlanganisa amakhono amaningana e-multi-agent: Ama-ejenti awuhlobo lwengxenye edume kakhulu onjiniyela abalufaka ekugelezeni nge-Langflow.

Ukugoqa: Ukusuka ku-copilot kuya ku-Pilot

Ukugeleza komsebenzi we-Agentic kuhlanganisa ndawonye idatha yebhizinisi, i-AI, nama-API, kwakha izinhlelo zokuzenzakalela ezinikeza amandla ochwepheshe besizinda ukuze balinganise amakhono abo futhi benze amabhizinisi asebenze kangcono nge-AI. Ukuhlanganisa ama-agent e-AI ekwakhiweni kwebhizinisi kuphawula ukweqa okukhulu endleleni izinhlangano ezibhekana ngayo nezinqubo ezizenzakalelayo nezamabhizinisi.

Lawa ma-ejenti, anikezwe amandla ama-LLM nezinhlaka ze-ejenti, eqa imingcele evamile ngokusebenza ngaphandle komthungo kuzo zonke izinqubo, ukugeleza komsebenzi, nekhodi.

Ukwamukela ukugeleza komsebenzi we-ejenti kuthembisa ukuthuthukisa ukusebenza kahle, ukuqina, kanye nokusabela kuyo yonke imisebenzi yebhizinisi. Bazophatha konke ukugeleza komsebenzi, baphathe imisebenzi eyinkimbinkimbi ngokuguquguquka okukhulu, futhi bathuthukise ukuzizwisa kwamakhasimende ngokunikeza ukusebenzisana komuntu siqu nokufika ngesikhathi. Njengoba i-automation ishumeka ezinhlelweni zebhizinisi, abashayeli bezindiza be-AI bazophothula izifundo zabashayeli bezindiza, futhi izinhlangano ezisebenzisa i-AI ye-agent zizoba sesimweni esingcono sokusungula izinto ezintsha, ukuncintisana, kanye nokuletha inani.

Ukuze uthole imininingwane eyengeziwe nge-agent AI, funda iphepha elimhlophe lamahhala, " Amasistimu Okuzenzakalela: Ikusasa Le-Enterprise Architecture I-Agentic. ” Futhi hlola leli khasi ukuze ufunde kabanzi mayelana ne-Langflow .

Ibhalwe ngu-Ed Anuff, isikhulu esiphezulu semikhiqizo, i-DataStax