目錄編程體驗的變化趨勢202320242025理解當前項目全項目理解能力生成實施計劃代碼生成代碼解釋

內(nèi)嵌聊天理解代碼庫20232024代碼伴侶2025主動協(xié)助者智能體Agent(/test)自動執(zhí)行生成單元測試的端到端過程測試覆蓋率更高，交付代碼更可靠使用內(nèi)置檢查保持控制權(quán)單智能體編程:單元測試生成掃描代碼，發(fā)現(xiàn)隱藏漏洞和質(zhì)量問題。近乎實時地生成修復建議滿足企業(yè)“安全左移”的需求單智能體編程:安全掃描&

代碼質(zhì)量Q

Detector

Library-支撐安全和代碼質(zhì)量掃描Agent基于多年的亞馬遜安全標準，“Security

isJob

Zero”長期維護，及時更新，基于規(guī)則和機器學習

的掃描規(guī)則在大多數(shù)流行編程語言的檢測方面，性能優(yōu)于領(lǐng)先的公開基準測試工具智能體Agent–Java項目現(xiàn)代化構(gòu)建&測試Java8或者11應用知識庫構(gòu)建&測試Java17/21使用生成式AI進行錯誤修復持續(xù)維護Javav8

/

v11JavaV17/21編程體驗的變化趨勢20232024代碼伴侶單智能體編程2025主動協(xié)助者智能體通過交互式編碼體驗自主執(zhí)行復雜工作流程通過自然語言對話和上下文感知優(yōu)化開發(fā)工作流程讀寫本地文件寫代碼跑測試調(diào)試問題部署代碼2024:代碼輔助,提升效率，提升質(zhì)量2025:與專家一起獨立完成人類需要數(shù)小時的工作2027:突破，AI能完成非常具備挑戰(zhàn)性的任務，需要團隊數(shù)年完成的任務。來源:[1]

Yes,you

can

measure

software

developer

productivity.McKinsey

&

Company,2023“使用標準AWS圖標創(chuàng)建一個橫向布局的基于AWS的簡單的電商平臺架構(gòu)圖，使用drawio格式保存。架構(gòu)圖包含用戶、負載均衡層、應用服務層（用戶、商品、訂單、購物車等服務on

EC2）、數(shù)據(jù)層（MySQL和Redis

on

EC2）、數(shù)據(jù)分析層（EMR

onEC2和S3數(shù)據(jù)湖）”Amazon

Q

Developer理解架構(gòu)圖生成IaC總結(jié)輸??…Amazon

Q

Developer

CLIAmazon

Q

DeveloperAmazon

CodeCatalystAI

Agent

in

PullRequest/describe/review/improveAmazon

BedrockAmazon

Q

Developer

CLIAmazon

Q

Developer

CLIQ

DeveloperCLIcontextQ

DeveloperCLIQ此貢獻者不會RustDeveloper

CLIQ

Developer

2427RoughideaImplementationRequirementsClarificationDesignSolutionImplementationPlanResearchCodebaseVibe

Coding(氛圍編程)Prompt-driven

development

(PDD)實際上，絕大部分的需求都來自于一個“rough

idea”29RoughIdeaImplementationRequirementsClarificationDesignSolutionImplementationPlanResearchCodebase30What

is

this

codebase?Do

more

research

on

the

q

chat

implementationandgive

me

asummaryCreate

a

directory

called

planning

and

write

thissummary

information

to

a

file

called

codebase.mdwrite

a

file

in

the

planning

folder

calledslashcommand.md

covering

how

slash

commands

areimplemented.

Include

enough

detail

so

a

developercould

implement

a

new

slash

command

using

thisinformation3132RoughIdeaImplementationRequirementsClarificationDesignSolutionImplementationPlanResearchCodebase33Read

rough-idea.md.Ask

me

one

questionat

a

time

so

we

candevelop

a

thorough,step-by-stepspec

forthis

idea...34RoughIdeaImplementationRequirementsClarificationDesignSolutionImplementationPlanResearchCodebase35Based

on

all

the

planningdocs

so

far,

create

adetailed

design

document

soa

developer

can

immediatelybegin

implementation.36RoughIdeaImplementationRequirementsClarificationDesignSolutionImplementationPlanResearchCodebase37Convert

theimplementation

plan

intoa

series

of

prompts

fora

code-generation

LLMthat

will

implement

eachstep

in

atest-drivenmanner...38RoughIdeaImplementationRequirementsClarificationDesignSolutionImplementationPlanResearchCodebase39Implement

prompt

1

fromthe

prompt

plan.

Updatetodo.md

with

progress./aws/amazon-q-developer-cli/pull/834生成式AI

下的新型軟件開發(fā)范式40Vibe

Coding(氛圍編程)v.s.Prompt-drivendevelopmentPrototype

?Production

readydevelopment？遵循軟件設(shè)計與開發(fā)規(guī)范跳過所有中間過程當Agent編程成為主流時，程序員的技能需求將發(fā)生根本性轉(zhuǎn)變。傳統(tǒng)必備的 編程語言、算法和數(shù)據(jù)結(jié)構(gòu)能力可能讓位于新的核心技能：精準的需求定義能 力，