Mfeɛ mmienu a atwam no, me ne me kuo a ɛwɔ Work & Co no asɔ ahwɛ na yɛde nkakrankakra de AI coding nnwinnadeɛ te sɛ Copilot, Cursor, Claude, ne ChatGPT abɔ mu de aboa yɛn ama yɛde wɛb osuahu a nnipadɔm de di dwuma no akɔ. Nokwarem no, wɔ mfiase no adwenem naayɛ bi ne aha bere kakraa bi akyi no, AI nnwinnade ahorow anya kwan akɔ me da biara da dwumadie mu. Bere kɔɔ so no, application ahorow a wɔahyehyɛ wɔ baabi a yehui sɛ ntease wom sɛ yɛbɛma AI agye tumi no fii ase yɛɛ kɛse, enti misii gyinae sɛ mɛkyɛ nsɛm bi a mfaso wɔ so a wɔde di dwuma ama AI nnwinnade ama nea mefrɛ no “asɛyɛde a ɔde di dwuma” no. Dɛn na mepɛ sɛ meka sɛ developer a ɔwɔ asɛyɛde? Ɛsɛ sɛ yɛhwɛ sɛ yɛde code a ɛyɛ papa bɛma sɛdeɛ yɛn a wɔdi dwuma ne yɛn afɛfoɔ hwɛ kwan no. Ɛnsɛ sɛ yɛn ntoboa (i.e., twe adesrɛ) bɛyɛ adesoa ma yɛn mfɛfo adwumayɛfo a ɛsɛ sɛ wɔsan hwɛ yɛn adwuma no mu na wɔsɔ hwɛ. Afei nso, sɛ ɛba sɛ woyɛ adwuma ma adwumakuw bi a: Ɛsɛ sɛ yɛn adwumawura pene nnwinnade a yɛde di dwuma no so. Ɛsɛ sɛ wodi nneɛma a ɛho hia te sɛ ahobammɔ ne kokoam nsɛm ho dwuma yiye: Mfa ahintasɛm, adetɔfo data (PII), anaa proprietary code nhyɛ nnwinnade mu a wunnyaa nhyehyɛe no kwan. Fa no sɛ code a efi ɔhɔho bi hɔ wɔ intanɛt so. Sɔ hwɛ na hwɛ sɛ ɛyɛ nokware bere nyinaa. Hyɛ no nsow: Saa asɛm yi fa no sɛ nimdeɛ bi a ɛyɛ mfitiaseɛ paa wɔ AI coding nnwinnadeɛ te sɛ Copilot wɔ VSCode anaa Cursor mu. Sɛ eyi nyinaa te sɛ nea ɛyɛ foforo koraa na wunnim a, Github Copilot video nkyerɛkyerɛ no betumi ayɛ mfiase a ɛyɛ nwonwa ama wo.
AI Coding Nnwinnade a Wɔde Di Dwuma a Ɛboa Hyɛ no nsow: Nhwɛsoɔ a ɛdidi soɔ yi bɛtwe adwene asi adwumayɛ so titire wɔ JavaScript-based wɛb aplikeshɔn te sɛ React, Vue, Svelte, anaa Angular. Nteaseɛ a Wobɛnya Wɔ Codebase a Wonnim Ho Ɛnyɛ nea ɛntaa nsi sɛ wobɛyɛ adwuma wɔ codebase a wɔde asi hɔ so, na sɛ wobɛka codebase kɛse bi a ɛyɛ agyapade ho a, ebetumi ayɛ hu. Bue wo project ne wo AI agent (wɔ me fam no, Copilot Chat wɔ VSCode mu) kɛkɛ na fi ase bisa nsɛm sɛnea wobɛbisa wo yɔnko dwumayɛni no ara pɛ. Mpɛn pii no, m’ani gye ho sɛ me ne AI dwumayɛni biara bɛkasa sɛnea mɛkasa akyerɛ me yɔnko onipa no ara pɛ. Nhwɛso a wɔayɛ no yiye kɛse a wɔde hwehwɛ nsɛm a wɔka kyerɛ ni: "Ma me architecture overview a ɛkorɔn: entrypoints, routing, auth, data layer, build tooling. Afei kyerɛw fael 5 a ɛsɛ sɛ wokenkan nnidiso nnidiso. Fa nkyerɛkyerɛmu ahorow sɛ hypotheses na si so dua denam jumping to referenced files so."
Wubetumi akɔ so abisa nsɛm a edi akyi te sɛ “Ɔkwan bɛn so na routing no yɛ adwuma kɔ akyiri?” anaa “Ka me kasa fa nokwaredi nhyehyɛe ne akwan horow no so” na ɛde wo bɛkɔ akwankyerɛ a ɛboa so de ahyerɛn hann bi akɔ esum mu wɔ codebase a wunnim mu. Triaging Breaking Nsakrae Bere a Woreyɛ Dependencies no Foforo Npm packages a wobɛsesa, titiriw bere a ɛba ne nsakrae a ɛbubu a, ebetumi ayɛ adwuma a ɛyɛ ɔbrɛ na egye bere, na ama woayɛ debug a ɛfata regressions. Nnansa yi ara na ɛsɛ sɛ mema data visualization library plotly.js no kɔ soro kɛse biako a wɔayi no adi no fi version 2 kɔ 3, na nea efii mu bae ne sɛ, axis labeling a ɛwɔ graphs no bi mu no gyaee adwumayɛ. Mekɔɔ so bisaa ChatGPT sɛ: "Meyɛɛ me Angular project a ɛde Plotly di dwuma no foforo. Meyɛɛ plotly.js — dist package no foforo fi version 2.35.2 kɔɔ 3.1.0 — na afei labels a ɛwɔ x ne y axis no so no ayera. Dɛn na esii?"
Ɔnanmusifo no san bae de ano aduru bi bae ntɛm ara (hwɛ w’ankasa wɔ ase hɔ). Hyɛ no nsow: Meda so ara di nkyerɛkyerɛmu no ho adanse tia aban atutra akwankyerɛ no ansa na mede fix no remena.
Refactors a Wɔreyɛ no Dwoodwoo wɔ Fael ahorow so Akyinnye biara nni ho sɛ codebases a ɛrenya nkɔso no da hokwan ahorow a ɛwɔ hɔ ma code consolidation adi. Sɛ nhwɛso no, wuhu sɛ code duplication wɔ fael ahorow a wobetumi ayi afi mu akɔ function anaa component biako mu. Nea ɛde aba ne sɛ, wusi gyinae sɛ wobɛbɔ ade a wɔakyɛ a wobetumi de aka ho mmom na woayɛ saa refactor no wɔ fael biako mu. Afei, sɛ́ anka wode nsa bɛyɛ saa nsakrae no wɔ wo fael a aka no mu no, wosrɛ wo agent no sɛ ɔmfa refactor no nkɔma wo. Agents ma wopaw fael ahorow pii sɛ nsɛm a ɛfa ho. Sɛ wɔyɛ refactor ma fael baako wie a, metumi de fael a wɔasan ayɛ ne nea wɔanka no nyinaa aka nsɛm a ɛfa ho no mu na maka akyerɛ agent no sɛ ɔmfa nsakraeɛ no nkɔ fael foforɔ te sɛ yei so: “Replicate the changes I made in file A to file B as well”. Nneɛma a Wɔde Di Dwuma Wɔ Mfiridwuma Ahorow a Wonnim Mu Aha-bere a m’ani gye ho paa a mede AI coding nnwinnade di dwuma no mu biako ne bere a ɛboaa me ma meyɛɛ animated gradient animation a ɛyɛ den yiye wɔ GLSL mu, kasa a na minnim no yiye. Wɔ nnansa yi adwuma bi mu no, yɛn adwumfo no de animated gradient bae sɛ loading state wɔ 3D ade so. M’ani gyee adwene no ho ankasa na na mepɛ sɛ mede biribi soronko na ɛyɛ anigye ma yɛn afɛfo no. Noɔhaw: Na mewɔ nna mmienu pɛ sɛ mede bedi dwuma, na GLSL wɔ adesua kwan a ɛyɛ den koraa. Bio, AI adwinnade bi (wɔ eyi mu no, ChatGPT) baa mfaso, na mifii ase kanyan no ara kwa sɛ ɔnyɛ HTML fael a egyina hɔ ma me a ɛkyerɛ canvas ne animated color gradient a ɛyɛ mmerɛw yiye. Anamɔn akyi anammɔn, me kanyan AI no sɛ wɔmfa finesse pii nka ho kosi sɛ miduu nea ɛfata a ebefi mu aba so sɛnea ɛbɛyɛ a metumi afi ase de shader no ahyɛ me codebase ankasa mu. Nea efii mu bae awiei koraa: Yɛn afɛfo ani gyei kɛse, na yɛde ade a ɛyɛ den mae wɔ bere tiaa bi mu esiane AI nti. Akyerɛw Sɔhwɛ Wɔ me suahu mu no, ɛntaa mma sɛ bere a ɛdɔɔso wɔ nnwuma mu a ɛbɛkɔ so akyerɛw na wɔahwɛ unit ne integration sɔhwɛ ahorow a ɛfata so, na ɛno so no, developers pii ani nnye adwuma a ɛne sɛ wɔbɛkyerɛw sɔhwɛ ahorow no ho ankasa. Wo AI boafoɔ a wobɛkanyan no ma wahyehyɛ na wakyerɛw sɔhwɛ ama wo no yɛ yie koraa na wobɛtumi ayɛ wɔ berɛ tiawa bi mu. Nokwarem no, ɛsɛ sɛ wo, sɛ́ ɔyɛfo no, da so ara hwɛ hu sɛ wo sɔhwɛ ahorow no hwɛ wo application no afã horow a ɛho hia no ankasa na wodi sɔhwɛ nnyinasosɛm ahorow a ntease wom akyi, nanso wubetumi “de” sɔhwɛ ahorow no nkyerɛwee no ama yɛn AI boafo no. Nhwɛso a wɔde ka asɛm: "Twerɛ unit tests ma saa dwumadie yi denam Jest so. Kata anigyeɛ kwan, edge cases, ne failure modes so. Kyerɛkyerɛ nea enti a test biara wɔ hɔ mu."
Wubetumi mpo de sɔhwɛ guru Kent C. Dodds’ sɔhwɛ ho nneyɛe pa no atwam sɛ akwankyerɛ ama wo nanmusifo, te sɛ nea ɛwɔ ase ha yi:
Nnwinnade a Wɔde Di Dwuma wɔ Mu Ɛte sɛ shader nhwɛsoɔ a yɛadi kan aka ho asɛm no kakra no, nnansa yi ara wɔmaa me adwuma sɛ menhwehwɛ code duplication mu wɔ codebase mu na memfa ntoto ho ansa na refactor bi aba ne akyi. Akyinnye biara nni ho sɛ ɛnyɛ adwuma a ɛho nhia sɛ wompɛ sɛ wobɛfa ɔkwan a egye bere pii a ɛne sɛ wode nsa bɛtoto fael ahorow ho no so a. Ɛnam Copilot mmoa so yɛɛ script bi a ɛhwehwɛɛ code duplication mu maa me, hyehyɛɛ na ɛhyɛɛ output no wɔ table bi mu, na ɛde kɔɔ Excel. Afei mefaa no anammɔn biako kɔɔ akyiri. Bere a wɔyɛɛ yɛn code refactor no, meka kyerɛɛ agent no sɛ ɔmfa me Excel krataa a ɛwɔ hɔ dedaw no sɛ mfitiasede, mfa nka ho wɔ mprempren tebea a ɛwɔ duplication mu wɔ column ahorow a ɛsono emu biara mu, na ɔmmu delta no ho akontaa. Updating Code a Wɔakyerɛw Bere Tenten a Abɛsen Biara Nnansa yi ara, m’afɛfoɔ dedaw bi bɔɔ me, ɛfiri sɛ berɛ kɔɔ so no, na nneɛma kakraa bi nnyɛ adwuma yie wɔ ne wɛbsaet no so bio. Nea wɔkyeree no: Ɛkame ayɛ sɛ wɔyɛɛ wɛbsaet no mfe du a atwam ni, na na JavaScript ne SCSS no de compile nnwinnade dedaw mmom te sɛ requireJS redi dwuma, na na nhyehyɛe no hwehwɛ sɛ wonya Node.js dedaw a ɛrenkɔ so mpo wɔ me 2025 MacBook so. Sɛ mede nsa yɛ ɔdansi nhyehyɛe no nyinaa foforo a, anka ebegye me nna pii, enti misii gyinae sɛ mɛhyɛ AI agent no sɛ, “So wubetumi ayɛ JS ne SCSS adansi nhyehyɛe no foforo akɔ lean 2025 stack te sɛ Vite?” Ɛyɛ ampa sɛ ɛyɛɛ saa, na bere a me ne agent no yɛɛ refining bɛyɛ dɔnhwerew biako akyi no, memaa me SCSS ne JS build no danee kɔɔ Vite so, na mitumi de m’adwene sii bugfixing ankasa so. Hwɛ hu ara sɛ wobɛma output ne compiled fael ahorow no ayɛ nokware yiye bere a woreyɛ nsakrae titiriw a ɛte saa wɔ wo build nhyehyɛe no mu no. Nsɛm a Wɔaboaboa Ano Na Wɔakyerɛw Wopɛ sɛ wobɔ wo mmara mu nsakraeɛ a woayɛ nnansa yi nyinaa mua wɔ kasamu baako mu ma commit nkrasɛm, anaasɛ wowɔ commits no din tenten na wopɛ sɛ wobɔ no mua wɔ bullet points mmiɛnsa mu? Ɔhaw biara nni ho, ma AI no nhwɛ so, nanso yɛsrɛ wo hwɛ hu sɛ wobɛkenkan no yiye. Nhwɛsoɔ a wɔde hyɛ obi so no yɛ mmerɛ te sɛ nkrasɛm a wode bɛmena wo yɔnko onipa: “Mesrɛ wo, bɔ nsakraeɛ a mayɛ nnansa yi wɔ atuo a ɛyɛ tiawa mu no mua”. M’afotuo wɔ ha bɛyɛ sɛ wode GPT bedi dwuma de akyerɛw ahwɛyiye, na sɛnea ɛte wɔ koodu ho no, yɛsrɛ wo hwɛ nea efi mu ba no ansa na wode amena anaa wode amena. Nyansahyɛ Ne Nneyɛe Pa Nkannyan a ɛkanyan Mfaso biako a ɛnyɛ nea ɛda adi kɛse wɔ AI a wode bedi dwuma so ne sɛ dodow a wo nsɛm a wɔka kyerɛ no yɛ pɔtee na wɔayɛ no sɛnea ɛfata no, dodow no ara na nea efi mu ba no ye. Adeyɛ a wɔde kanyan AI dwumayɛni bi hyɛ yɛn ma yɛhyehyɛ yɛn ahwehwɛde ahorow pɔtee sɛnea yebetumi ansa na yɛakyerɛw na yɛakyerɛw mmara. Eyi nti na, sɛnea mmara kyerɛ no, mekamfo kyerɛ kɛse sɛ fa wo nkannyan no yɛ pɔtee sɛnea wubetumi. Ryan Florence, a ɔne no kyerɛw Remix no, de ɔkwan a ɛyɛ mmerɛw nanso tumi wom a wobɛfa so ama saa adeyɛ yi atu mpɔn denam kasamu a wode bewie wo mfiase no so: “Ansa na yɛbɛhyɛ aseɛ no, wowɔ nsɛmmisa bi ma me?”
Saa bere yi de, AI no taa san ba de nsɛmmisa a ɛboa ba a wubetumi ama w’adwene pɔtee no mu ada hɔ, na akyerɛ ɔnanmusifo no kwan ma wama wo ɔkwan a ɛfata kɛse ama w’adwuma no.
Fa Version Control Di Dwuma Na Yɛ Adwuma Wɔ Digestible Chunks Mu Version control a wode bedi dwuma te sɛ git a ɛnyɛ sɛ ɛba mfaso bere a woreyɛ adwuma abom sɛ kuw wɔ codebase biako so nko na mmom sɛ ɛbɛma wo sɛankorankoro a ɔde mmoa ma a ɔwɔ nsɛntitiriw a ɛyɛ pintinn a ɛsɛ sɛ ɔsan kɔ so sɛ tebea a egye ntɛmpɛ si a. Esiane sɛ ɛnyɛ nea wɔahyɛ da ahyɛ da nti, ɛtɔ mmere bi a AI tumi yɛ ɔkwasea na ɛyɛ nsakrae a ɛnyɛ mmoa kɛkɛ mma nea worebɔ mmɔden sɛ wubenya no na awiei koraa no abubu nneɛma a wontumi nsiesie. Sɛ wokyekyɛ w’adwuma mu kɔ commits pii mu a, ɛbɛboa wo ma woabɔ nsɛntitiriw a ɛyɛ den a wobɛtumi asan akɔ so sɛ ɛba sɛ nneɛma kɔ nkyɛn a. Na wo mfɛfo kuw no nso bɛda wo ase, efisɛ ɛbɛyɛ mmerɛw sɛ wɔbɛsan ahwɛ wo mmara no mu bere a wɔakyekyɛ mu ayɛ no nsensanee a wɔahyehyɛ no yiye wɔ ntease mu no. Hwɛ mu Yie Eyi yɛ adeyɛ a eye sen biara a ɛkɔ so wɔ ɔkwan a ɛkɔ akyiri so, nanso wɔ m’adwene mu no, ɛbɛyɛ nea ɛho hia kɛse mpo bere a wode AI nnwinnade redi dwuma ama nkɔso adwuma no: Yɛ wo mmara no mu nhwehwɛmufo a odi kan. Hwɛ sɛ wobɛgye bere kakra akɔ wo nsakraeɛ no so nkyerɛwdeɛ biara, sɛdeɛ wobɛhwɛ obi foforɔ mmara mu no, na wode w’adwuma no bɛmena bere a atwam w’ankasa wo ho nhwehwɛmu no nkutoo. “Nneɛma abien nyinaa yɛ nokware ma me mprempren: AI adwumayɛfo yɛ nwonwa na ɛma adwumayɛ kɔ soro kɛse. Sɛ wudum w’amemene na wugyae koraa a, ɛyɛ mfiri akɛse a wɔde bɔ nneɛma nso.”— Armin Ronacher wɔ ne blog post Agent Psychosis: Are We Going Insane?
Awiei Ne Nsusuwii a Ɛho Hia Wɔ m’adwene mu no, AI coding nnwinnade betumi ama yɛn adwumayɛ atu mpɔn sɛ developers da biara da na ama adwene mu tumi ahofadi ama nhyehyɛe pii ne nsusuwii a ɛkorɔn. Wɔhyɛ yɛn ma yɛka nea yɛpɛ sɛ yebefi mu aba no ho asɛm kɔ akyiri. Ɛtɔ mmere bi a, AI biara betumi ayɛ nsusuwii hunu, a nea ɛkyerɛ titiriw ne sɛ ɛda ahotoso nne mu. Enti yɛsrɛ sɛ hwɛ hu sɛ wobɛhwɛ na woasɔ ahwɛ, titiriw bere a w’adwene mu yɛ wo naa no. AI nyɛ dwetɛ tuo, na migye di sɛ, nea ɛkyɛn so ne tumi a wotumi siesie ɔhaw ahorow sɛ ɔdebɔfo no renkɔ akyiri da. Wɔ developers a wɔrefi ase wɔ wɔn adwuma mu nkyɛe fam no saa nnwinnade yi betumi ayɛ sɔhwɛ kɛse sɛ wɔbɛyɛ adwuma no fã kɛse no ara ama wɔn. Nea ebia ɛbɛyera wɔ ha ne adwuma a ɛtaa yɛ draining na ɛyɛ yaw denam bugs ne nsɛm a ɛyɛ anifere sɛ wobɛsiesie na wɔadi ho dwuma, aka “the grind”. Cursor AI’s ankasa Lee Robinson mpo gye eyi ho kyim wɔ ne nsɛm a ɔde too gua no biako mu:
AI coding nnwinnade renya nkɔso ntɛmntɛm, na m’ani agye nea ɛbɛba akyi no ho. Mewɔ anidaso sɛ wuhui sɛ asɛm yi ne n’afotu no boa wo na w’ani agye sɛ w’ankasa wobɛsɔ eyinom bi ahwɛ.
