Izixhobo zokuKhowuda ze-AI zisekwinqanaba le-R&D

Ngokutsho kwe -Stack Overflow ye-survey ye-2024 , i-76% yabaphuhlisi basebenzisa okanye baceba ukusebenzisa izixhobo ze-AI-ngoku bayingxenye yomsebenzi. Bancedisa kwimisebenzi eqhelekileyo kodwa banokucaphukisa xa bevelisa ubuvuvu ngokuzithemba. Ngelixa i-YouTubers yakha "ukuqala kwebhiliyoni yeedola" kunye ne-ChatGPT ekhuthazayo kunye neearhente ze-AI zithatha umhlaba veki nganye, amaqela okwenyani asacinga ukuba angazisebenzisa njani ezi zixhobo ngokufanelekileyo. Namhlanje, ukuqonda uncedo lwe-AI kubaluleke kakhulu njengokwenza ikhowudi okanye uyilo lwenkqubo-kufuneka siziqhelanise kwaye sikhawuleze.

Ingxaki ezi zixhobo zisekwinqanaba le-R&D-zitshintsha rhoqo, zikopana, kwaye zifanelwe yikhredithi ngokusombulula iingxaki ebezingasonjululwanga ngaphambili. Bonke abanazo izikhokelo zokusebenzisa ezicacileyo. Nditsho noCopilot, ngaphandle kokusekwa ngakumbi, akanazo izifundo ezicacileyo kunye nezona zenzo zilungileyo. Isicombululo? Siza kwenza eyona nto bayenzayo abaphuhlisi-silungelelanise kwaye senze isakhelo ngokwethu.

Umtshintshi womdlalo

...kwaye kwakhona " I-Quantum leap "," i-Paradigm shift "," ii-Agents ziyeza ", uyinike igama. Ngelixa ezi zixhobo ngokwenene ziguqula ukuhamba kwethu komsebenzi, utshintsho lusebenza ngakumbi: abaphuhlisi ngoku basebenza njengabakhokheli beqela, ukulawula abancedisi be-AI endaweni yokubhala ikhowudi ngokuthe ngqo. Izakhono ezingundoqo zitshintshele ekuyileni, ekucwangciseni, ekuchazeni nasekuphononongeni.

Iikhonsepthi eziphambili ze-UX ezi zixhobo zazisayo zezi:

1. Iingcebiso ezingaphakathi - eyona ndlela yendalo yokudibanisa. Xa zisebenza, azinamthungo-kodwa xa ziphazamisa ikhowudi yakho, kukuqhawuka kwangoko kwimveliso.
2. Ncokola -indlela yakudala apho iiLLMs zinxibelelana nekhowudi yesiseko sakho ngokusebenzisa imiyalelo ekhokelwayo
3. Umqambi (" Copilot Hlela "," Cline act ") - ingqiqo entsha abaninzi bayifumanisa ibhidekile . Kukutshintshela kwiiarhente ezizimeleyo. Umqambi usebenza kwiifayile ezininzi kwaye unokusebenzisa utshintsho kwaye aphindaphinde ngokuzenzekelayo ngokusekwe kwiimpazamo, iingxaki zokubhala njl.
4. Iiarhente -indaleko elandelayo elindelweyo-ezizimeleyo ngokupheleleyo kunye nabancedisi be-AI abadityaniswe ngokuthe ngqo kwi-IDE yakho.

Ukuqhelana " neengcebiso " akukho nzima. Abancedisi be-AI baqala nabo, nto leyo eyathi yabamba ingqalelo yethu kuqala. Iincoko zithe ngqo: faka iifayile kumxholo, phinda, sebenzisa, kwaye uqinisekise iziphumo.

Izixhobo zodidi lomqambi zinika imiceli mngeni eyongezelelekileyo ukuze zikwazi kakuhle, zifuna igophe lokufunda kunye neendlela ezingacacanga. Okwangoku, umhleli we-Cursor unikezela ngesona sixhobo "somqambi" esifikelelekayo, ngelixa u-Copilot elandela ngokusondeleyo ngo- " Copilot Edits ," esanda kwazisa ukuhamba komsebenzi okusekwe kwiarhente.

Ukuze ube nolwazi ngabaqambi, kufuneka uqonde iikhonsepthi ezintathu eziphambili:

1. Imiyalelo
2. Imigaqo
3. Umxholo

Makhe sihlolisise nganye kwezi.

Imiyalelo

Njengeqela elikhokelayo kunokuba nje abaphuhlisi, kufuneka siqale nayiphi na iprojekthi entsha okanye into enkulu ngokudala uXwebhu loYilo okanye uXwebhu olucacileyo lweeMfuno zeMveliso . Olu qheliselo luphuhlisa ubunjineli obuqinileyo kunye nokucinga kwemveliso ngelixa kugcinwa ixesha elininzi lokuphumeza. Eyona nto ingcono kukuba la maxwebhu anokuba:

1. Yenziwe nge-AI
2. Kwaye emva koko isetyenziswe njengemiyalelo yabaqambi

Ukwenza la maxwebhu, qala uqokelele iimfuno ebantwini , emva koko ujonge imodeli yokuqiqa kwiNgxoxo. Zombini i-Copilot kunye ne-Cursor zineemodeli zokuqiqa ezakhelwe ngaphakathi ezifanelekileyo kulo msebenzi. I-OpenAI's o1 kunye ne o3-mini zifumaneka ngokungagqibekanga, ngelixa i-Cursor's Chat ixhasa i-DeepSeek-R1 (nangona ingekabikho kumqambi wayo ) - zonke izixhobo ezibalaseleyo zale njongo.

Isiqhelo esilungileyo kukugcina amaxwebhu oyilo kwinqanaba eliphezulu lerepo yakho (siya kusebenzisa ifolda requirements ) eququzelelwe ziimpawu, kunye ProjectOverview.md kwingcambu. Nanku umzekelo wesakhiwo kwiimfuno zewebhu ye-Twitter:

 requirements/ ├── ProjectOverview.md # Core product description └── Features/ ├── Authentication.md # User registration ├── Tweet.md # Tweet CRUD ├── UserProfile.md # Profile management ├── Engagement.md # Likes, retweets ├── Infrastructure.md # Storage, caching, etc └── ...

Ukuba yonke into imiswe ngokufanelekileyo, ukongeza uxwebhu loyilo lwenqaku elitsha kulula njengokubhala olu ncedo:

Ukugcina imiyalelo kwi-codebase yakho kunika iingenelo ezicacileyo: ulawulo lwenguqulelo, ukugcinwa lula, kunye nokuhamba okusemgangathweni kwePR. Nangona kunjalo, amalungu eqela angengobuchwephesha abanje ngabanini beMveliso, abaphathi, kunye nabayili be-UX banokufuna ukufikelela ngaphandle kokusebenzisa i-git. Nazi izisombululo ezithathu:

1. Gcina yonke into kwi-Notion, upapashe amaphepha omyalelo, kwaye uwafake njengamaxwebhu usebenzisa @Docs indlela emfutshane.

2. Yenza umbhobho oguqula amaphepha eNgcaciso kwiifayile .md kwaye uzigcine kwindawo yokugcina
3. Fundisa iqela lakho ukusebenzisa i-git-elona khetho liluncedo kwiqela lonke

Nje ukuba imiyalelo yakho ifikeleleke kumhleli wakho, tshintshela kumqambi kwaye uqalise ukuphumeza. Oku kusikhokelela ekubeni silungelelanise iMithetho .

Imigaqo

Okwangoku, kuphela i-Cursor exhasa " imigaqo " - imiyalelo yokuphunyezwa ngokuthe ngqo kwiifayile/iifolda ezithile. Eli nqaku liza kunwenwela kwabanye abahleli, kuquka i-VSCode Copilot, okwangoku enikezela kuphela " kwiifayile ezikhawulezayo " ezingenako ukudityaniswa ngokuthe ngqo kwi-codebase.

Imithetho yeCcursor ibanzi ngakumbi - cinga CONTRIBUTING.md idityaniswe nemithetho ye-linter kwaye iphuculwe yi-LLM. Le migaqo yimveliso-agnostic, inokwabelana, kwaye idlulise ngokufanelekileyo ulwazi, izenzo ezingcono, kunye neenkcukacha zokuphunyezwa kumaqela kunye nabasebenzisi bethala leencwadi.

Imithetho ingenziwa ngephalethi yomyalelo kwaye igcinwa kwifolda yeprojekthi yakho .cursor/rules enolwandiso lwe .mdc . Le fomati inika amandla amanqaku aphezulu afana ne- @mentioning iifayile ezithile kwikhowudi yakho. Kucetyiswa kakhulu ukuba uzibophelele le mithetho kwindawo yakho yokugcina kwaye usebenzisane ekuyiphuculeni. Nantsi indlela yokusebenza yokusebenzisa imithetho:

1. Imithetho yekhesa yophando ngokuthe ngqo kwisitaki sakho setekhnoloji, iqala ngoluhlu olugcinwe njengeereferensi. Umzekelo, ungafumana imithetho yekhesa ebhalwe kakuhle ye Next.js kunye ne React esebenza njenge templates ezilungileyo.
2. Hlaziya imithetho ngokuqhubekayo ngexesha lophuhliso. Xa ubona ipateni enokuthi yenziwe ibe ngumgaqo ngelixa ubhala ikhowudi, yibhale ngokukhawuleza kwifayile yakho yemigaqo.
3. Funda kwezona zibalaseleyo entsimini. Indlela entsha yokuba abaqulunqi bethala leencwadi babelane ngolwazi kunye nokwandisa ukuthathwa komntwana kukudala imithetho ekhethekileyo yabancedisi be-AI. Ndiyazi ngeenkampani ezimbalwa ezenza oku - iConvex igqame ngokwenza imithetho yazo zombini iimodeli ze-OpenAI kunye ne-Anthropic kunye nokwabelana ngazo kumaxwebhu azo . Ngelixa ndingasebenzisanga imveliso yabo, ukugxila kwabo ekuphuculeni amava omphuhlisi ngokudityaniswa kwe-AI kuyanyanzeleka. I-Supabase ngomnye umzekelo omhle .

Amathala eencwadi amaninzi afuna ngokungxamisekileyo imithetho ye-AI. Ukusuka kwimbono yomphuhlisi we-frontend, ndingaxhamla ngokuba nazo kwi -TanStack Query , i-React Spring , i-Firebase kunye nezinye ezininzi-ezininzi. Le migaqo ingagcina ixesha elibalulekileyo kwaye incede ukuthintela iimpazamo eziqhelekileyo ezenziwa ngabaphuhlisi xa befunda ubugcisa obutsha.

Umxholo

Khumbula ukubandakanya zonke iimeko ezifanelekileyo - okukhona unikezela ngedatha esemgangathweni, kokukhona uya kufumana iziphumo ezingcono. Umhleli wekhesa une-advanteji ngaphezulu kwe-Copilot apha ngokuvumela iindidi ezininzi zomxholo:

1. Uxwebhu- lusebenza kakuhle ngenene, lunike nje indawo yokungena kulo naluphi na uxwebhu, luya kulukhuphela, lucazulule, kwaye lugcinele iimfuno zexesha elizayo.
2. Uphendlo lwewebhu - hayi ngokungqongqo umxholo, lubonelela ngokukhawuleza kwimithombo ye-intanethi
3. Izixhobo ezahlukeneyo zophuhliso-i-git ethile yokwenza, iimpazamo ze-lint, ii-Notepads, kunye nezinye izinto zakudala.
4. Iiseva ze-MCP zinokubonelela ngemeko yexesha langempela. Nangona ukuseta kuluqili kancinci, axabisekile xa ufuna ufikelelo lwedatha ephilayo.

Emva kokuqonda ezi zixhobo, inyathelo lakho elilandelayo kukuphucula ukusebenza komntu ngamnye kunye neqela. Kodwa yeyiphi indlela eya phambili ukusuka apha?

UCline kunye neRoo-Code. Ulawulo

Uyakuhlala ujongene norhwebo phakathi kokulula kunye nolawulo, phakathi kwezisombululo ezizenzekelayo kunye nokwenziwa kwezigqibo ngesandla. Ukuba uzimisele ukuntywila nzulu kwaye awoyiki ukujongana neempazamo, imingeni yokusebenza, kunye nemiphetho erhabaxa, cinga ngokuzama iCline (okanye ifolokhwe yayo yeRoo-Code , enefilosofi eyahluke kancinane).

Zombini izixhobo ziyilelwe ukubonelela ngocoselelo olukhulu kwinto eyenzekayo phantsi kwe-hood kangangoko kunokwenzeka:

1. Ziyimithombo evulekileyo kunye nobhaliso-simahla. Endaweni yoko, usebenzisa ezakho izitshixo ze-LLM API okanye iinkonzo ezifana ne -OpenRouter , uhlawula kuphela okusebenzisayo.
2. UCline ubonisa ngokucacileyo yonke imisebenzi yayo, kuquka ukuba zeziphi iifayile ezifundayo nezizilungisayo.
3. I-Cline inikezela ngezimvo ezineenkcukacha kunxibelelwano lwe-LLM, ubume befestile yemeko, kunye neendleko zeseshoni yengxoxo nganye.
4. Ibonisa iindlela zoCwangciso/zoMthetho eziqondayo - indlela enengqiqo ekufuneka ezinye izixhobo ziqwalasele ukuyamkela.

Into ebulalayo kukuba uCline unokuqhuba kwaye alungise isicelo sakho - yinyani kwaye iyasebenza, njengoko uza kubona xa uyizama.

Ukuba konke oku kunomdla kuwe, jonga inqaku lakutshanje lika-Addy Osmani , elibonelela ngentshayelelo egqwesileyo kwaba bahleli.

Ukuqukumbela

Ukwamkela ezi zixhobo akulona uhambo olulula, kwaye ungalindelanga ukubhala "iprojekthi yonke ukusuka ekuqaleni phantsi kwemizuzu emi-5." Nangona kunjalo, le yindlela ecacileyo eya phambili.

Itekhnoloji sele ikhona, kodwa siphoswa kukuhamba komsebenzi okudityanisiweyo kwe-AI eququzelela lonke iqela- ingengabo abaphuhlisi kuphela, kodwa abaphathi nabaqulunqi - bajikeleze ezi zixhobo zitsha. I-AI inokuziva isoyikeka, kwaye ukwabelana ngempembelelo yayo kunokubonakala kungakhululeki ekuqaleni (njengokuxelela iqela lakho elikhokelayo ukuba i-AI ibhale i-80% yenqaku ngoqwalaselo olucokisekileyo). Nangona kunjalo, uphuhliso lwesoftware luya kuvela kuphela xa ezi zixhobo ziba yinto ehambelana nokuhamba komsebenzi weqela. Olona tshintsho luphumeleleyo lwenzeka kumaqela akhuthaza iingxoxo ezivulelekileyo malunga namava e-AI, ukuphononongwa kwesixhobo esisebenzisanayo, kunye nokufaka igalelo ngenkuthalo ezona zenzo zilungileyo ezifundiweyo kuluntu olubanzi lophuhliso.