I-slop - lokhu ingozi etholakalayo ngokuvamile lapho sincoma imikhiqizo ye-AI. I-something ephakeme, ezinzima ukulawula, eyakhelwe ngempumelelo kwamanani efanelekayo nge-akhawunti eyodwa: ukujabulela ukujabulela, ukujabulela isikhathi, ukukhiqiza i-hype. Umbala wayo, kanye nemiphumela yayo yokushintshwa kwe-potencial, akuyona. bubble ku-industry Konke lokhu ngokushesha ukhuthaza i-debat mayelana nemiphumela enhle kanye ne-danger ye-AI. Ubuchwepheshe ye-software akuyona isizukulwane. Ucwaningo ibonisa ngokuvamile ukuthi ukufundisa izixhobo ze-AI akuyona elula, futhi ngaphandle kwezobuchwepheshe ezinhle, imiphumela yokusebenzisa kwabo kungenzeka ngisho . Ukuphendula Nangona kunjalo, i-AI ingasiza kakhulu ekuthuthukiseni isofthiwe. Yonke okufakiwe kuyinto ukucindezeleka, ukucindezeleka, ukucindezeleka. Noma ngokuhambisana ne-checklist. I-1: Thola i-motivation Ukuthuthukiswa kwe-software akuyona ngokwemvelo eminyakeni. Sine-methodology, izinguquko zokusebenza, izindlela zokuthumela, ukucaciswa, QA, njll. Kodwa-ke sinikeza ikhodi umugqa ngesilinganiso, njenge-tekst, ukugcina ukulawula ukuthi sinikeza njengoba siphinde. Lezi zinyanga Ngithanda noma kungekho, ikhodi ngeke ikhiqizwa ngokushesha ikakhulukazi ngama-model ye-language. Lokhu kubuyekezwa kakhulu. Siye kufuneka ukuguqulwa kwezinto ezininzi - ngokuvamile ukuguqulwa kwebhizinisi lethu. Lokhu kunezinto emangalisayo kakhulu. Ngokuvamile, izizathu ezinzima kakhulu zokusebenzisa imishini ye-AI ekuthuthukiseni ngosuku zonke zihlanganisa: Ngaphansi 
 
 
 Kuyinto enhle. Zonke izikhathi ngifunde ukwenza I-AI ukusebenza ngempumelelo kwami, ngifanele ngokugcwele. Kuyinto ukujabulela isixhobo entsha, ukufundisa, futhi umsebenzi eyenziwe kahle. Kuyinto engatholakali. Isikhokelo enhle kakhulu, kodwa omunye enhle kakhulu. Ufuna ukubona inkalo futhi uzobe ufumane ngokushesha. Ukukhishwa kwe-AI kuyinto enhle manje futhi uzobe kuyindlela enhle elandelayo. It is becoming a job requirement. Umzekelo enhle iyona le standard yokuthengisa kusuka : Ukuhlobisa Ukubuyekeza ukuba unayo i-motivation efanelekayo kuyinto elula - nje ukhangele izibonelelo zakho lapho izinguquko ezintsha ze-LLM kanye nemikhiqizo ye-AI yokukhiqiza isofthiwe. I-emotion efanelekayo kuyinto ukujabulela ukuthenga isixhobo entsha enhle noma enhle kakhulu. Uma, emzimbeni nokumangalisayo, uzothola ukucindezeleka, kungenzeka ukuthi ungaphendula kakhulu kumadivayisi amasha. Kodwa ukucindezeleka kungaba futhi - usebenzise ukucindezeleka ukufundisa izixhobo ezintsha. Ukukhuthaza ukuthi ungathanda lapho okuhloswe ukucindezeleka kwenziwa isizathu esilandelayo esikhulu ukucindezeleka. Ukukhiqiza 2. Isilinganiso Ngemuva kwalokho, zonke izihloko ze-checklist zihlanganisa njenge- . Umgca lapha kulula - okungenani, ngokufanelekileyo, futhi okwengeziwe iminyango LLM inikeza, okungcono izinga lokukhipha. Ubuchwepheshe I-context specificity iyatholakala ngokusebenzisa isakhiwo se-codebase efanelekayo - ukunciphisa izinto ezingenalutho, amamoduli eyahlukile nge-spectrum esifanele ye-responsibility. Indlela elula yokuhlanganisa izindandatho ezivela ku-monorepos. Namhlanje, akufanele kakhulu ukulungiselela i-repository ephelele ye-multi-language yeprojekthi jikelele. Ukuqala ngokuvumelana nezinhlangano zangaphakathi - izicelo ezisebenzayo kanye nezinhlangano ezahlukahlukene ezahlukahlukene ezisebenzayo nge-utility code. Izinsizakalo zihlukile - izicelo kufanele zihlukile iziphakheji, futhi iziphakheji zihlukile ngamunye, kodwa iziphakheji zihlukile iziphakheji. Nazi isakhiwo isibonelo ungasebenzisa. Kuyinto efanelekayo kakhulu izimo: root/
├── applications/
│   ├── website/
│   ├── public-api/
│   ├── admin-dashboard/
│   └── ...
│
├── packages/
│   ├── js/
│   │   ├── ui/
│   │   ├── client/
│   │   ├── validation/
│   │   └── ...
│   │
│   ├── py/
│   │   ├── common-lib/
│   │   ├── models/
│   │   ├── auth/
│   │   └── ...
│   │
│   └── go/
│       ├── common/
│       ├── clients/
│       └── ...
└── Makefile
 I-Makefile iyasebenza njenge-enterpoint ye-everything ukuze ikwazi ukuhanjiswa kwe-script ye-tech agnostic. I-Python ne-Go inikeza i-monorepos ngokuvamile ngaphandle kwebhizinisi. Ukuze i-JS, i-pnpm iye yakhelwe ukhetho olungcono - namhlanje ngithole nge-npm kanye ne-wire. Uhlobo le-monorepo iyinhlangano elula kakhulu, elula, futhi ifakwe. Futhi isebenza kahle kakhulu nge-AI agents. Ukukhishwa kwe-package kulula njengoba ikhodi iyatholakala kakhulu futhi inikeza umsebenzi esifanele. Uma usebenza nge-code ku-applications, ungakwazi: 
 
 
 
 Ukuhlola zonke izipakheji ze-external ezijulile Yenza izinguquko ngezinyathelo, ukuguqulwa amapakheji okokuqala, bese isigaba esisodwa se-application Ukuguqulwa kwama-error e-isolation, isinyathelo esinyathelo, ngaphakathi kwe-feedback loop. Lokhu kuphela inesibopho yayo. 3. Setup Feedback Loop I-Quality ye-AI-generated code iyahambisa kakhulu uma umphathi we-AI angasebenza izivivinyo, ama-linters, futhi usebenzisa i-output kusuka kumadivayisi ezisetshenziselwa ukuguqulwa kweziphumo ngesikhathi sokusebenza. Ngo-ideal world, isixazululo engcono kuyinto ukunikeza I-AI ukufinyelela ku-code execution environment - ngoba i-web projects, okungenani ama-browsers. Ukukhula kwezinto ezivela ngokushesha, kodwa manje kunazo . akukho isixazululo esifanele Ngaphezu kwalokho, ngisho izindiza zokufaka kanye nezinhlelo zokusebenza zokusebenza zindlela ezinhle zokulawula ikhwalithi nokunciphisa i-hallucinations. Ikakhulukazi ngenxa yokufaka ukuhlolwa kwe-code ye-static ku-monorepo kuyinto engabizi kakhulu. Futhi ungathuthukisa izindiza ze-unity ngokusekelwe ku-code eyenziwe. Uma ezinye izixhobo ze-AI (njenge-Cursor) zihlola linters ngokushesha, kunzima ukufundisa ngokucophelela i-LLM ukuhlola izivakashi. Ngokwesibonelo, i-prompt ingaba kuhle: Update packages/js/api-client:

Add getUserProfile(userId) method:
- GET /users/{userId}/profile
- Include tests for success and 404 error

Run `make test-api-client` and fix any failures.
 I-Memory-Bank yokwakha I-Memory-bank iyisisombululo esifundisa indlela yokuhlanza nokulawula izinhlobo ezimbili ze-entities: Waze . Rules Documentation Kuyinto enhle kakhulu - kuyinto imikhiqizo yakho ebizwa kusukela izibuyekezo ezahlukene: imikhiqizo, ukwakhiwa, kanye nemikhiqizo izinga. Kukhona imikhiqizo eminingi online mayelana nokubhala imikhiqizo yobuchwepheshe enhle. Yini kubaluleke kwethu kuyinto ukuthi imikhiqizo enhle kakhulu futhi kulandipha ikhodi, kulula ukuba i-agent ukufinyelela. Documentation Umgangatho we-inthanethi ye-inthanethi ye-agent ye-AGENTS.md ifayela. Lesi ifayela kufuneka ufakele isakhiwo se-projekthi kanye nenkqubo, futhi inikeza izixhumanisi kwamanye amafayela zebhizinisi. A monorepo setting makes it convenient to organize documentation within specific modules while alsoining a root-level memory bank. Ngiyavuza kanjani isakhiwo sefayela se-memory-bank: monorepo/
├── memory-bank/                          # Shared across entire repo
│   ├── AGENTS.md                         # Root entry point
│   ├── monorepo-conventions.md           # Overall structure & rules
│   ├── shared-tech-stack.md              # Common technologies used
│   ├── development-workflow.md           # Git, CI/CD, deployment
│   └── architecture-overview.md          # High-level system design
│
├── applications/
│   ├── website/
│   │   ├── memory-bank/
│   │   │   ├── AGENTS.md                 # References root + specific context
│   │   │   ├── product-overview.md       # What this app does
│   │   │   ├── user-journeys.md          # Key user flows
│   │   │   ├── tech-stack.md             # Specific technologies
│   │   │   ├── api-integration.md        # How it connects to APIs
│   │   │   └── tasks.md                  # Current work items
│   │   └── src/
 I-Rule is a relatively new concept and, in my opinion, one of the most productive ideas for reusing prompts. I-Rule is encoding guidelines written in natural language — essentially a combination of Waze . Umhlahlandlela we-Cursor ku-Rules kubonakala ikakhulukazi enhle, njengokukwazi ukuxhumana nezinsizakalo ezivamile ezivamile ezisebenzisa izakhiwo ze-globe. Lokhu ngokuvamile "ukuthumela" izinsizakalo ezivamile ku-context lapho usebenza ngezinye izingxenye ze-repository yakho. Ungathola izinsizakalo ze-TypeScript, Supabase, noma GitHub Actions nje, bese kusetshenziswe ngeziprojekthi. Lokhu kubalulekile ikakhulukazi lapho usebenza ngezindlela ze-naming ne-repository structure standards. Rules contribution.md .eslintrc Izinsizakalo zihlanganisa futhi zihlanganisa ngokushesha. I-Best Practice kuyinto ukuvikelwa ku-root repository - lokhu kubalulekile kubhalisa izidakamizwa ze-code. I-"Free Page" problem ye-Documentation kanye ne-Rules iyakwazi ukuvikelwa kalula. I-Documentation ingasebenza nge-AI nge-induction elula bese kusetshenziswe. Futhi izinhlelo ze-technologies ezithile zitholakala kalula ku-intanethi: Ngena ngemva Ngena ngemva . Ukukhanya ikhaya imikhiqizo Ngokwenza lokhu, zonke izivakashi nge-AI agent zihlanganisa ngokufunda i-Memory-bank futhi zihlanganisa ukuhlaziywa. Lokhu kuncike ukucindezeleka futhi ukucindezeleka, kodwa kubhalwe ngokushesha njengoba kusiza isikhathi lapho kuqala isivakashi esilandelayo. I-Learn “Plan → Act” Pipeline Umphumela wePlan → Act inokukwazi ukucubungula kakhulu ekusebenzeni ne-AI assistants, futhi kuyinto indlela eyodwa enokutholakalayo ukuhlangabezana nezinsizakalo ezinzima. Izinzuzo eziningi ze-AI encoding ziye zihlanganisa okuzenzakalelayo. I-Cursor, isibonelo, i-automatically plans the task execution without requiring additional instructions. Kodwa ukuze uthole imiphumela enhle, ikakhulukazi ngezinsizakalo ezinkulu, kungcono ukuqhuba i-Plan → Act ngesandla. Isisombululo esekelwe ku-90/10 isisombululo: Ukulungele ukuchithwa kwe-90% yesikhathi sakho futhi kuphela i-10% ekukhiqizeni. I-Planing ikhiqiza izakhiwo ezithile — imisebenzi. Isisombululo se-best practice kuyinto ukugcina lezi zokusebenza e-memory bank e-module-level, e-.md file. Izifundo zakho zihlanganisa, zihlanganisa, futhi zihlanganisa, kakhulu iziphumo zakho zihlanganisa. Umhlahlandlela we-Plan → Act Workflow kuhle: Ukulungiswa 
 
 
 
 
 
 
 Ukubuyekezwa kwe-context. Ukubuyekeza izidakamizwa ezidingekayo kusuka ku-memory bank Ukucaciswa kwe-business context. Ngokuvamile isitimela se-user for the feature Ukucaciswa kwe-Technical Specification - Okufakiwe QA umklamo inqubo - okuyinto scripts ukuqhuma ukuhlolwa, okuyinto ukuhlolwa ukongeza Ukwakha ifayela tasks.md I-Iterating nge-tasks.md ifayela, izibuyekezo, izincazelo, izincazelo Ngena ngemvume 
 
 
 
 
 
 
 Ukukhishwa kwe-session entsha (i-chat, i-tab) Ukuqala ukuvelisa. Step by step, ngokulandelana Ukubuyekeza imiphumela ngokuvamile kanye ne-tests ngalinye iminyango. Ukuhambisa ngemuva kwezinye iminyango. Ukuhlobisa i-feedback loop, ukongeza ukubuyekezwa kwe-test Ukuhlaziywa kweDokumentation Ukususa ifayela tasks.md ikhaya Ukuze usebenzisa le pipe: important tips 
 
 
 Ukuhlukanisa njalo i-Plan ne-Act ku-sessions ezahlukile. Kuyinto elula, futhi kuncike ukuzinza nokuphephisa ama-tokens Uma i-agent inikeza imiphumela embalwa, akufanele imiyalezo ezengeziwe - bheka isinyathelo sePlan, ukunciphisa task.md yakho, memory-bank, futhi izinhlelo. Ukubuyekezwa kwePlan → Act Pipeline akuyona kuphela ku-AI literacy. Kuyinto enhle, ngisho enhle, ingcindezi. Ngokuvamile, i-Plan enhle kuyinto idokhumenti ye-design eyenziwe ngempumelelo — ungakwazi ukuxhumana ne-team yakho noma i-team yakho, ukulungiselela ngaphambi kokusebenza ku-PR, futhi usebenzisa njengoba isisekelo sokuphepha nokuhlolwa kwakho. I-Plug-in Cloud I-Agent ye-Plug-in Kukho izixhobo eziningi ezahlukene emakethe, futhi angathanda ukucindezeleka kwelinye eyodwa esifanele. Kodwa siphinde izigaba ezimbili eziyinhloko: izesekeli zendawo (njenge-Claude Code, Cursor, ne-Codex CLI) kanye nezixhobo ze-cloud (njenge-Codex Cloud, Jules, noma i-Newly Released Ngathi Claude Code ku-website Njengomthengisi, uzokufinyelela izixhobo ezisekelwe ama-agents ezivamile. Noma i-AI-powered code editor noma i-CLI tool, izinhlelo eziningi zokusebenza zokusebenza zokusebenza zokusebenza zokusebenza ngokushesha futhi ngokushesha. Kodwa izixhobo ezisekelwe ku-cloud zihlanganisa izindawo zayo zayo zayo. Into yokuqala kuyinto ukuthi zonke ukuxhumana kwenziwa nge-browser - akufanele izinga le-developer noma inqubo esisekelwe ngendawo. Umthamo wesibili yesibili kuyinto ukuthi ziye ziye ziye ziye ziye ziye ziye ziye zihlanganisa imiphumela eminingi ngokuvamile. Ngenxa yalokho, zikhona ezinhle ngezibonelo zokusetshenziswa: 
 
 
 
 Noma i-codebase analysis, i-time estimate ye-implementation, i-evaluation ye-potential difficulties, noma i-work plan creation (task.md). Lolu hlobo lwe-analysis iyisiza kakhulu ngaphambi kokufaka umsebenzi olukhulu. Uyazixazulule ngokushesha ukuthi izinhlayiya ze-plan kungenziwa ngokuvumelana ne-AI, okuvimbela isikhathi sokuvumelela ngaphezulu. I-refactoring ye-scale-scale kodwa e-refactoring e-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring ye-refactoring. Ukushintshwa kwe-non-critical, okushintshwa kwe-team ye-developer. Ngaphandle kokwenza umsebenzi ku-Task Tracker, i-PO yakho noma i-Designer nge-Codex ukufinyelela kungenziwa i-command ye-cloud agent. Emva kwalokho uxhumane imiphumela ngokushesha. I-principal thing is keeping the tests green. Izixhobo olungcono iyinhlangano enhle, futhi okungenani, inqubo yokusungula izinhlangano efanayo. I-cloud encoding agent ingangena omunye kwezi zokusungula. 7. Ukuhlobisa Ngaba ingxenye enhle ye-AI encoding iyinkimbinkimbi. Ungayifaka kanjani izindlela ezinhle? Ungayifaka i-workflows phakathi kwama-teammates, futhi wonke umuntu kufanele usebenzisa isixhobo esifanayo? Ngiyaxolisa imibuzo enhle kakhulu. Izinzuzo ze-AI zibonisa izinzuzo eziningana kakhulu: kanjani ukulawula ibhanki le-memory futhi umuntu kufanele inesibophelele ukucubungula? Indlela yokusebenza nge-PRs? Indlela yokuxhumana nezinsizakalo ze-AI phakathi kwabanye? Umphakeli we-AI isixazululo ukufinyelela okuhle noma okuhle, kodwa akukho izixazululo ezivamile. Njengesikhathi esivamile ezivela ku-software engineering, sizothintela ukufundisa isixazululo. Ngokusho yami, indlela enhle yokusabalalisa ulwazi kunazo nge-forums ezivela ezingenalutho. Uma unayo izivakashi ezivamile ze-developer yokusabalalisa izimfuneko, ukuxhumana nezindaba, kanye nokudlulisa izinzuzo, ukwenza ukuchithwa kwe-AI isikhwama ebalulekile kulezi zihlanganisa. Uma unayo isikhwama elifanayo, mhlawumbi kunezinto ukwenza isikhwama, noma ukhangela ukuchithwa kwe-AI ku-plattform efanayo.

This story contains new, firsthand information uncovered by the writer.

Read My Stories

Lo msindo ukhiqizwa ngolimi lokuqala lwendaba!

Hlola le Slop. Hlola Ukubhalisa Ngcono nge-AI

About Author

IMIBONO

HANG TAGS

LESI SIHLOKO SETHULWE NGAPHAKATHI

Related Stories

The Billionaires F*cked Up

Zulu Network: Moving The Bitcoin Economy Forward With a Two-Tiered Bitcoin Layer 2 Architecture

The Billionaires F*cked Up

Zulu Network: Moving The Bitcoin Economy Forward With a Two-Tiered Bitcoin Layer 2 Architecture

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps