9 Amathuluzi Angcono Kakhulu we-AI Okuthuthukiswa Okushayelwa Ngokukhethekile ngo-2026: I-Kiro, i-BMAD, i-GSD, kanye nokunye Qhathanisa
Njengoba ama-ejenti wokufaka amakhodi we-AI ekhula enekhono, kuvele inkinga yesakhiwo: isivinini ngaphandle kokucaca. Onjiniyela bakhiqiza ikhodi yokusebenza ngemizuzu, kuphela ukuthola izinsuku kamuva ukuthi ayifani nalokho okudingwa yisistimu. I-Spec-driven Development (SDD) ibhekana nalokhu ngokuqondile – ngokuphatha ukucaciswa okuhlelekile njengomthombo weqiniso nekhodi njengokuphuma kwakho okukhiqiziwe, kunenye indlela.
Lolu hlu luhlanganisa amathuluzi ayi-9 e-AI asetshenziswa abathuthukisi empeleni ukusebenzisa ukuhamba komsebenzi kwe-SDD ngo-2026.
AWS Kiro
🔗 kiro.dev | Amadokhumenti | Amamodeli
I-Kiro i-IDE ye-agent eyakhiwe eduze nentuthuko eqhutshwa okuthile, edizayinelwe ukuthatha onjiniyela ukusuka kumqondo baye ekukhiqizeni ngokuqina okuhlelekile esikhundleni sokuphindaphinda. Kunokuba ibhale ikhodi futhi icele i-AI ukuthi isize endleleni, i-Kiro idinga ukuthi onjiniyela benze ngokusemthethweni inhloso kuqala. Ibaqondisa kunqubo yezigaba ezintathu — Izimfuneko, Idizayini, kanye Nemisebenzi — ikhiqiza ama-artifact amathathu ahlelekile: needs.md, design.md, and tasks.md. Imininingwane yobuchwepheshe ephawulekayo: I-Kiro ikhiqiza izindaba zomsebenzisi isebenzisa isaziso se-EARS (I-Easy Approach to Requirements Syntax), ekhiqiza umbandela wokwamukela ohleliwe ohlanganisa amacala asemaphethelweni onjiniyela abangawasingatha mathupha.
Umehluko omkhulu uhlelo lwayo lwamahhuku e-ejenti – ama-automation aqhutshwa umcimbi avutha lapho amafayela alondolozwa noma edalwa, ukuphatha imisebenzi efana nezibuyekezo zokuhlola, ukuvuselela kwe-README, nokuskena kwezokuphepha ngaphandle kokwaziswa okwenziwa mathupha. Ngokukhethwa kwemodeli, okuzenzakalelayo kwe-Kiro irutha ye-Auto ehlanganisa amamodeli emingcele eminingi – okuhlanganisa u-Claude Sonnet, Qwen, DeepSeek, GLM, kanye ne-MiniMax – futhi ikhetha imodeli efanelekile ngomsebenzi ngamunye ukuze ibhalansise ikhwalithi nezindleko. Onjiniyela bangaphinda baphine imodeli ethile yokuziphatha okungaguquki. Yakhelwe ku-Code OSS, abasebenzisi be-VS Code bazozizwa besekhaya ngokushesha. I-Kiro iphinde isekele i-CLI nesixhumi esibonakalayo sewebhu, futhi ayidingi i-akhawunti ye-AWS ukuze isetshenziswe. Kuhle kakhulu emaqenjini adinga ukugeleza komsebenzi okuhlelekile endaweni yokuthuthukiswa ejwayelekile.
I-GitHub Spec Kit
🔗 github.com/github/spec-kit | Okuthunyelwe Kwebhulogi
I-GitHub Spec Kit iyinketho yomthombo ovulekile eyamukelwa kakhulu ngumphakathi ekuthuthukisweni okushayelwa ngokukhethekile – i-Python CLI enezinkanyezi ezingu-93,000+, ukukhishwa kwakamuva kube yi-v0.8.7 (May 7, 2026), esekela ama-30+ AI ama-ejenti okubhala amakhodi okuhlanganisa i-Claude Code, i-GitHub Copilot, i-Amazon Q, ne-Gemini. Ukugeleza komsebenzi kuhamba ngezigaba ezine ezinezindawo zokuhlola ezicacile: Cacisa (ithwebula umongo webhizinisi kanye nemibandela yempumelelo), Hlela (ihumusha imininingwane ibe izinqumo zezakhiwo), Imisebenzi (ichitha amapulani ibe amayunithi ahlolekayo, abukezekayo), kanye Nokusebenza (isebenzisa ama-ejenti e-AI ngaphansi kwalezo zingqinamba).
Esisekelweni sakho konke ukuhamba komsebenzi kwe-Spec Kit “umthethosisekelo” — ifayela lemithetho yokubeka phansi eliqukethe izimiso ezingenakuguquleka zeleveli ephezulu ezisebenza kulo lonke ushintsho kuyo yonke isikhathi. Lokhu kuba yinkontileka eqhubekayo phakathi konjiniyela kanye nomenzeli. Ifilosofi ye-Spec Kit, njengoba i-GitHub yayizimele, ukuthi ikhodi manje isiwumphumela wemayela lokugcina: inhloso ingumthombo weqiniso, futhi ukucaciswa kuyasebenziseka. Yindawo yokuqala ezenzakalelayo yamaqembu amasha ku-SDD kanye nenketho ephathwayo kakhulu yamaqembu afuna ukugcina i-IDE yawo ekhona.
I-BMAD-INDLELA
🔗 github.com/bmad-code-org/BMAD-METHOD | Amadokhumenti
I-BMAD-METHOD (Yakha Amaphupho Ezakhiwo Eziningi) iwuhlaka lomthombo ovulekile olunelayisense ye-MIT oluhlela ama-agent ayi-12+ akhethekile e-AI kuwo wonke umjikelezo wokuphila wokuthuthukiswa kwesofthiwe. Inguqulo engu-6.6.0 yathunyelwa ngo-April 29, 2026, iphrojekthi yafinyelela kuzinkanyezi ze-GitHub ezingu-46,700+ kanye nezimfoloko ezingaphezu kuka-5,500. Ama-ejenti ayi-12+ ahlanganisa izindima ezihlukile ze-SDLC – okuhlanganisa ukuphathwa komkhiqizo, ukwakhiwa kwezakhiwo, i-UX, ukuthuthukiswa, i-QA, nemisebenzi ye-scrum master – futhi basebenzisane ngokunikezwa okuhleliwe okususelwe efayeleni: i-ejenti ngayinye ifunda umbhalo ophumayo we-ejenti yangaphambili futhi ibhale owakhe, igcine iketango elilandelekayo kusukela ezidingweni ngokudiliva.
I-V6 yethule Ithimba Le-Cross Platform Agent, evumela ukulungiselelwa okufanayo komenzeli ukuthi asebenze kuyo yonke Ikhodi ye-Claude, i-Cursor, i-Codex, nabanye abasingathi ngaphandle kokulungiswa kabusha. Isakhiwo se-V6 siphinde sihlukanise ukukhathazeka kube izendlalelo ezintathu: I-BMad Core (uhlaka lomhlaba wonke lokusebenzisana kwabantu ne-AI), Indlela ye-BMad (imojula yokuthuthukisa esheshayo eyakhelwe ku-Core), kanye ne-BMad Builder (evumela amaqembu ukuthi akhe futhi abelane ngama-ejenti angokwezifiso nokugeleza komsebenzi). I-BMAD iwuhlaka lokuya emaqenjini afuna ukugeleza komsebenzi okuhleleke kakhulu, okuhlukaniswe indima eminingi ngaphandle kokukhiya komthengisi. Uhlaka lumahhala ngokuphelele awekho ama-paywall.
Ikhodi yokwengeza
🔗 augmentcode.com | Umhlahlandlela we-SDD
I-Augment Code isondela ekuthuthukisweni okushayelwa okuthile kusukela kusendlalelo lomongo kunesendlalelo sokugunyaza esicacisiwe. I-Context Engine igcina ukuqonda okuqhubekayo kwezakhiwo kuwo wonke amafayela angu-400,000+ – ibhekana negebe lomongo we-cross-repository eliphula ukugeleza komsebenzi okucaciswe kakhulu esikalini, ikakhulukazi kuma-codebases we-brownfield yezinsizakalo eziningi. I-Augment ibika u-70.6% kubhentshi le-SWE (uma kuqhathaniswa nesilinganiso semboni esingu-54%) kanye ne-59% ye-F-score kubhentshimakhi yokubuyekezwa kwekhodi ye-AI; lezi zibalo zibikwe ngabathengisi futhi kufanele ziphathwe ngokufanele.
Imodeli yayo ye-BYOA (Letha Umenzeli Wakho) ivumela amaqembu ukuthi axhume Ikhodi ye-Claude, i-Codex, noma i-OpenCode eduze nomenzeli wayo we-Auggie. I-Augment Code ayibhali ama-specs ngokwemvelo – amaqembu asadinga ithuluzi elifana ne-Spec Kit noma i-Kiro lokuphathwa kwe-spec ehlelekile – kodwa inikeza isisekelo se-semantic esenza lezo zibalo zinembe kuwo wonke ama-codebase amakhulu. Ifaneleka kakhulu emaqenjini ebhizinisi asebenzisa izakhiwo eziyinkimbinkimbi zezinsizakalo eziningi lapho ukukhukhuleka komongo, hhayi ukudalwa okucacisiwe, kuyimodi yokwehluleka okuyinhloko.
Claude Ikhodi
🔗 claude.ai/code | Amadokhumenti
I-Claude Code iyithuluzi lomugqa womyalo we-Anthropic, futhi ngokungafani namathuluzi afana ne-Cursor noma i-GitHub Copilot engeza ukuhamba komsebenzi kanjiniyela, yakhelwe ukuthuthukiswa okuzibusa ngokugcwele – ukuhlela, ukuhlela ukuhamba komsebenzi okunezinyathelo eziningi, nokubuza imibuzo yokulandelela ngaphandle kokwaziswa njalo. Ngokugeleza komsebenzi okushayelwa yi-spec, i-Claude Code iphatha imibhalo emikhulu yezicaciso kahle phakathi neseshini eyodwa, icubungula amasethi ezimfuneko eziphelele futhi ikhiqize ukusetshenziswa ngephasi eyodwa ehambisanayo.
Onjiniyela bavamise ukusebenzisa amafayela e-CLAUDE.md njengesendlalelo sokucaciswa — indlela engasindi ephoqelela umongo wephrojekthi eqhubekayo, amazinga okubhala amakhodi, kanye nemikhawulo yezakhiwo kuyo yonke isikhathi. Lokhu kusho ukuthi onjiniyela abaningi sebevele benza uhlobo lwe-SDD nge-Claude Code ngaphandle kokuyilebula kanjalo. Ikhodi ye-Claude iphinda isebenze njenge-ejenti yokusabisa evame ukusekelwa kuzo zonke izinhlaka ze-SDD ezihlanganisa i-BMAD, i-GSD, ne-GitHub Spec Kit.
I-GSD (Get Shit Done)
🔗 github.com/gsd-build/get-shit-done
I-GSD iwuhlaka oluqhutshwa yi-meta-prompting kanye nohlaka lobunjiniyela bomongo olwakhelwe ngokuyinhloko i-Claude Code kanye nama-ejenti ahambisanayo, ezibeka njengenye ye-BMAD enciphile, enemikhosi ephansi. Le phrojekthi idlule izinkanyezi ze-GitHub ezingu-61,000 – ikhula isuka ku-zero iye kuleso sibalo ngaphansi kwezinyanga ezinhlanu kusukela ekuzinikeleni kwayo kokuqala kukaDisemba 2025. Ifaka nge npx get-shit-done-cc@latest futhi isebenza kuyo yonke Ikhodi ye-Claude, i-OpenCode, i-Gemini CLI, iCodex, i-Copilot, i-Cursor, i-Windsurf, i-Augment, ne-Cline.
I-orchestration yayo enama-agent amaningi izala abacwaningi abafanayo, abahleli, ababulali, neziqinisekisi, ngayinye isebenza ewindini lomongo omusha elinamathokheni afika kwangu-200K anikezelwe ekusetshenzisweni. Idizayini ye-model-agnostic – ehlanganisa ukusekelwa kwe-OpenRouter namamodeli endawo – ihlukanisa ukuhamba komsebenzi kunoma yimuphi umthengisi we-LLM oyedwa. Lapho i-BMAD ingeza imicimbi ye-sprint kanye nokuxhumana nababambiqhaza, ifilosofi ye-GSD iwukuthi inkimbinkimbi kufanele iphile ohlelweni, hhayi ukuhamba komsebenzi. Iphinde igcwalise igebe i-Claude Code ngokwayo engalifaki ngokwendabuko: ukuzungezisa umongo, amasango ekhwalithi, nokuphikelela kwesimo sokuhlela kuzo zonke izikhathi.
Ikhesa (enemodi Yohlelo + Imithetho Yephrojekthi)
🔗 cursor.com | Imikhuba Engcono Kakhulu Yomenzeli
I-Cursor isalokhu ingomunye wabahleli be-AI abasetshenziswa kakhulu, futhi i-Plan Mode yayo ikwenza kube indawo esebenzayo yokungena kumaqembu asebenzisa imikhuba eqondile ngaphandle kokushintsha amaketanga amathuluzi. I-Plan Mode idala uhlelo oluningiliziwe lokuqalisa ngaphambi kokuba kubhalwe noma iyiphi ikhodi – ukubuza imibuzo ecacisayo, ukwenza imephu kwamafayela athintekile, nokwenza uhlelo olubuyekezekayo olugunyazwa unjiniyela ngaphambi kokuba umenzeli enze okuthile. Lokhu kuvimbela ukukhiqizwa kwekhodi ngaphambi kwesikhathi kuzici ezithinta amafayela amaningi noma ezidinga izinqumo zezakhiwo.
Ngomongo ofana nokucaciswa okuqhubekayo, isistimu yemithetho yamanje ye-Cursor isebenzisa imithetho yephrojekthi egcinwe ngaphansi kwayo .cursor/rules/ (omdala .cursorrules ingqungquthela manje isithathwa njengefa). Uma ihlanganiswe nemithetho yephrojekthi, i-Cursor isekela ukuhamba komsebenzi okungasindi, okuphathekayo okuphathekayo kuzici zenkundla eluhlaza okotshani okuphakathi nendawo kuya kokukhulu. I-tradeoff ukuthi ukusekelwa kwe-Cursor's spec akukona okwemvelo ekwakhiweni kwayo ngendlela i-Kiro eyiyo – awukho umjikelezo wempilo owakhelwe ngaphakathi, ukutholwa kokukhukhuleka, noma ukuvumelanisa kwe-living-spec. Emaqenjini afuna ukuthuthukiswa kwe-AI ehlelekile ngaphakathi komhleli ojwayelekile, wekhwalithi ephezulu ngaphandle kwe-SDD engaphezulu, Ikhesa Ngemodi Yohlelo iyindawo ephakathi enekhono.
I-OpenSpec
🔗 github.com/Fission-AI/OpenSpec
I-OpenSpec iqondise esimweni sokusetshenziswa esithile nesilinganiselwe: amaqembu lapho ukuphathwa koshintsho kudinga imibhalo ecacile, engahlolekayo ngaphambi kokuba noma yikuphi ukuqaliswa kuqale. Isebenzisa ukugeleza komsebenzi okumaphakathi kwesiphakamiso okunama-artifact ahleliwe ukuze kube nezinguquko, futhi ibhekana ngokukhethekile ne-brownfield iteration enamaka be-delta (ADDED/MODIFIED/REMOVED) alandelela ukuthi iluphi ushintsho oluhlobene nokusebenza okukhona kunezincazelo zenkundla eluhlaza. Okubalulekile, imibhalo ye-OpenSpec uqobo iyibeka njengento engasindi futhi evumelana nezimo esikhundleni sohlelo oluqinile lwesigaba – ihlinzeka ngesakhiwo ngaphandle kokuphoqelela amasango okugunyaza aqinile phakathi kwezigaba.
Ocwaningweni oluzimele lwangoFebhuwari 2026 lwenziwa ezigabeni eziyi-13 zamaphuzu ku-backend ye-Python yosayizi omaphakathi engenawo iseva, i-OpenSpec yathola amaphuzu aphezulu kakhulu sekukonke – nakuba izinga lishintsha kakhulu ngezinto eziza kuqala ezihlukile. Amaqembu ashintsha ukuziphendulela kwawo kanye nemizila yemibhalo edlula ukuvumelanisa okucacisiwe kokuphila azokuthola kufaneleka kakhulu. Ngemizamo emikhulu yezinsiza eziningi, ukumatanisa i-OpenSpec nenkundla ye-living-spec kuyatuswa, njengoba ukwakheka kwayo okusekelwe esiphakamisweni kukhiqiza amadokhumenti amile angakhukhuleka phakathi nokuqaliswa okunwetshiwe.
I-Tessl
🔗 tessl.io | Spec Registry | Amadokhumenti
I-Tessl iyinkundla yokunika amandla i-ejenti yolimi eyakhiwe ngemikhiqizo emibili ehlukene. I-Tessl Framework ifaka “njengamathayela” kumaphrojekthi .tessl/ lwemibhalo futhi ifundise noma iyiphi i-ejenti ehambisana ne-MCP – okuhlanganisa Ikhodi ka-Claude, Ikhesa, nabanye – ukulandela ukuhamba komsebenzi okushayelwa okuthile kungakhathaliseki isitaki: ama-ejenti abuza imibuzo ecacisayo kuqala, abhale imibhalo yezicaciso ehlelekile, alinde ukugunyazwa kukanjiniyela, bese eqalisa. Ama-Specs ahlala ku-codebase njengenkumbulo yesikhathi eside, enikeza izinqumo umkhondo wokucwaninga futhi evumela umenzeli ukuthi aguqule uhlelo lokusebenza ngokuhambisana ngokuhamba kwesikhathi.
I-Tessl Spec Registry iyisihlukanisi esicace kakhulu seplathifomu: ukubhaliswa okuvulekile okungaphezu kuka-10,000 okuchaza indlela yokusebenzisa kahle imitapo yolwazi yangaphandle yemithombo evulekile, eqondise ngokuqondile ukukhohliswa kwe-API nokuxubana kwenguqulo abasebenzeli abavame ukuyikhiqiza kumakhodi okukhiqiza. Kucabange njenge-npm ukuze uthole ukucaciswa – amaqembu afaka kokubili ithayela lendlela yokusebenza (indlela yokusebenza) kanye namathayela womtapo wezincwadi (maphi amathuluzi okufanele asetshenziswe ngendlela efanele) ukuze avimbele kokubili isiphithiphithi senqubo kanye nokubona izinto ezingekho ngaphansi kwemibhalo. Ukwakhiwa kwezendlalelo ezimbili – umongo wenqubo kanye nomxholo womtapo wezincwadi – ukuqonda okuyisisekelo kwe-Tessl: ukuhamba komsebenzi okuhlelekile kukodwa akwanele uma i-ejenti isabonisa ama-API akha ngawo.
Zizwe ukhululekile ukusilandela Twitter futhi ungakhohlwa ukujoyina wethu 150k+ ML SubReddit futhi Bhalisela ku Iphephandaba lethu. Linda! ukutelegram? manje ungasijoyina kuthelegramu futhi.
Udinga ukusebenzisana nathi ekuthuthukiseni i-GitHub Repo yakho NOMA Ikhasi Lobuso Lokugona NOMA Ukukhishwa Komkhiqizo NOMA I-Webinar njll.? Xhumana nathi
