AI 时代的程序员英语

12.1 AI 时代英语能力的重构

2023 年以来，"我用 ChatGPT 就够了"是程序员对英语学习最常见的退路。这个想法只对了一半：AI 确实能帮你消除 80% 的英语写作负担，但同时让剩下 20% 的能力变得更值钱。

这 20% 是什么？

1. 判断力：AI 给你三个版本，你能挑出哪个最自然。
2. 临场反应：会议上、面试中没时间问 AI。
3. 定制力：AI 默认输出"中等水平英文"，你要能让它输出"你想要的风格"。
4. 校对力：AI 也会犯错（特别是术语、语境、文化）。

这一章的目标：把第 1-11 章学到的英语，和 AI 工具有机结合。

12.2 让 AI 翻译/润色 commit message

Prompt 模板 1：commit message 润色

You are a senior engineer writing a Conventional Commits message.

Rewrite my Chinese description as a commit message following:
- Subject ≤ 50 characters
- Imperative mood (Add, not Added)
- Type: feat / fix / chore / refactor / perf / docs / test / build / ci
- Optional scope in parens
- Optional body (why, not what); separate from subject by blank line

Description:
"修复了用户在 1 秒内连点删除按钮会导致后端 500 错误的问题，
原因是前端没做防抖。"

Return only the commit message. No explanation.

典型输出：

fix(ui): debounce delete button to prevent duplicate requests

A rapid double-click on the delete button could fire two
backend requests within 1s, triggering a 500 due to the second
attempting to delete an already-removed resource. Added 300ms
debounce on the click handler.

12.3 让 AI 当 native speaker reviewer

Prompt 模板 2：英语母语审校

You are a native English speaker who is a principal engineer at
a Silicon Valley company. Review the following English text I
wrote for [PR description / blog / email].

Provide:
1. A revised version that sounds natural to a native speaker.
2. A bulleted list of specific changes you made and why.
3. Flag any expressions that came across as "non-native" or
   "AI-generated".

Be direct — I prefer accurate criticism over polite hedging.

Text:
"""
<粘贴你的英文>
"""

这个 prompt 的关键点：

• 明确角色和场景，让 AI 用对应语体。
• 要"具体改了什么 + 为什么"，不要只看输出——要学习。
• 要求"flag non-native"——把你的中式痕迹找出来。
• "direct" 防止 AI 客气敷衍。

12.4 让 AI 解释陌生英文

Prompt 模板 3：解释 RFC 段落

I'm reading RFC 7230 (HTTP/1.1). Explain the following paragraph
to me as if I'm a working engineer (not a protocol designer).
Translate jargon into concrete examples. Mention what real-world
implication this has for someone implementing or debugging an
HTTP server.

Paragraph:
"""
A sender MUST NOT send a message containing a Transfer-Encoding
header field unless the corresponding request indicates HTTP/1.1
(or later). A server MUST NOT send a Transfer-Encoding header
field in any 2xx (Successful) response to a CONNECT request.
"""

Prompt 模板 4：解释错误信息

I got the following error / stack trace. Explain:
1. What it literally means in plain English.
2. The most likely root causes (ranked by frequency).
3. The first 3 things I should check.
4. Bonus: if you've seen this exact error in a known library bug,
   mention which library and version.

Stack trace:
"""
<粘贴 stack trace>
"""

Environment: Node 20 / npm 10 / TypeScript 5.4 on macOS

Prompt 模板 5：解释论文摘要

Here is the abstract of a paper. Explain:
1. The problem in plain English (no jargon).
2. The key insight or technique the paper proposes.
3. What this would change in industry if proven.
4. What would be the simplest possible toy implementation.

Abstract:
"""
<粘贴 abstract>
"""

12.5 让 AI 帮你 review code review 评论

Prompt 模板 6：review 评论英语化

I'm leaving a code review comment on a coworker's PR. The
coworker is a [senior / junior / peer] from [country/culture].
Rewrite my draft to be:
- Direct enough to be useful
- Soft enough not to feel like an attack
- Conventional Comments labeled (praise / nitpick / suggestion /
  issue / question / thought / chore)

If my draft is borderline rude, flag it explicitly.

Draft:
"""
This is wrong. The cache key doesn't include the locale. Fix it.
"""

典型润色后：

issue: The cache key doesn't include the user's locale. Two users
with different locales requesting the same resource will share
the cache, which would return the wrong language. Could we add
locale to the key, or am I missing something about how this is
called?

12.6 让 AI 帮你写 Issue / PR 描述

Prompt 模板 7：从 git diff 生成 PR 描述

Here is the git diff for my PR. Generate a PR description
following this template:

## Why
- (problem statement)

## What
- (key changes)
- **Out of scope**: (things deliberately not done)

## How
- (architecture / approach)

## Test plan
- [ ] Manual verification step
- [ ] Unit test added

Closes #<issue>

Diff:
"""
<粘贴 git diff>
"""

Issue I'm fixing:
"""
<粘贴 issue 描述>
"""

Make the language match a senior engineer at a startup. No
"introduces", "leverages", or other AI-flavored words.

12.7 让 AI 帮你 mock 面试

Prompt 模板 8：Behavioral 面试模拟

You are a hiring manager at <Company> interviewing me for a
Senior Software Engineer role. Conduct a 30-minute behavioral
interview.

Rules:
- Ask one question at a time. Wait for my answer.
- Probe with follow-ups (especially "Tell me more about how you
  decided X").
- After each answer, give me a brief private rating (1-5) and
  one suggestion for what would make it stronger.
- Cover at least: conflict, failure, leadership, ambiguity.

When ready, ask the first question.

Prompt 模板 9：System Design 模拟

You are a Staff Engineer at <Company> conducting a 45-minute
System Design interview. Topic: design a URL shortener.

Rules:
- Start by asking what I want to clarify.
- Push back when my reasoning is hand-wavy.
- Ask follow-ups specifically about: scaling, reliability,
  data consistency, edge cases.
- Don't volunteer answers — make me earn each insight.
- After 45 minutes (estimated), give a rubric-based assessment.

Prompt 模板 10：think-aloud coding

I'm going to solve a LeetCode problem while thinking aloud in
English. Please:
- Listen as I explain my approach.
- Push back if my reasoning is wrong.
- Comment on whether my think-aloud is clear and structured.
- At the end, rate my "communication during coding" 1-5.

The problem is: <paste problem>

I'll start now.

12.8 让 AI 帮你写邮件

Prompt 模板 11：英文工作邮件

Write a professional but warm English email for the following
situation. Match the tone of a senior engineer at a US tech
company — confident, concise, friendly without being saccharine.

Situation: I want to follow up on a job offer I received yesterday.
I'm interested but want to negotiate base salary up by ~10%
because I have another competing offer at that level. I want to
keep the relationship warm.

Recipient: Recruiter named Sarah at the company.

Prompt 模板 12：拒绝 offer

Write a polite but warm decline email. The hiring manager spent
substantial time interviewing me; I want to leave the door open
for the future.

Reason for declining: I accepted a role at another company that
better aligns with my career goals (I don't want to specify the
company).

Prompt 模板 13：请求介绍

Write a cold email to <person> asking for a 15-minute coffee
chat / referral. The recipient and I share <mutual connection /
shared school / shared open source contribution>. Be specific
about what I want and respect for their time.

12.9 让 AI 帮你做技术博客

Prompt 模板 14：技术博客起草

I want to write a technical blog post about <topic>. Help me
draft an outline that:
- Opens with a Hook+Problem (a real story or surprising fact)
- Has 3-5 main sections, each with a clear takeaway
- Ends with a "what to try" section, not a generic conclusion
- Avoids: "in today's fast-paced world", "delve into",
  "leverage", "robust", "powerful", "navigate"

My audience: working engineers with 3-7 years experience.
My voice: direct, slightly opinionated, concrete examples.

Topic: <具体话题>

Prompt 模板 15：删除 AI 味

The following text reads like ChatGPT wrote it. Rewrite it to
sound like a human engineer who knows the topic well.

Hard rules:
- Cut every "Furthermore", "Moreover", "However", "In addition",
  "It's important to note that"
- Replace abstract claims with specific numbers / examples
- Vary sentence length — mix short punchy sentences with longer ones
- Add at least one opinion or strong claim

Text:
"""
<粘贴 AI 味文字>
"""

12.10 让 AI 给你定向训练

Prompt 模板 16：模拟 standup

You are my engineering manager. Conduct a 5-minute mock standup.

Rules:
- Greet the team. Ask each "person" (just me) for their update.
- After my update, ask one specific follow-up.
- If my update is too vague or too long, give feedback.
- Use 50-50 American / Indian English accents in your responses
  (just describe the accent, don't try to render it).

Start now.

Prompt 模板 17：检查我的 think-aloud 习惯

I'm going to describe how I solved a recent coding problem.
Listen and:
- Identify any vague phrases ("kind of", "stuff", "thing")
- Identify weak transitions ("then I did", "and then")
- Suggest stronger replacements
- Rate the structure: did I cover analysis → choice → implementation
  → test → complexity?

I'll start now: ...

Prompt 模板 18：找出我的 fillers

Here's a transcript of me speaking at a meeting. Find:
- All filler words (um, uh, like, you know, I mean, basically)
- All apology words I overused (sorry, just, maybe)
- Any sentences that started but never finished

Show me the count and the first 3 examples of each.

Transcript:
"""
<粘贴你的会议转录>
"""

12.11 实时字幕和翻译工具

主流工具对比

实战 setup 推荐

# 中-高频远程会议程序员的标配
1. 主屏：Zoom / Google Meet
2. 副屏：Otter.ai 实时转写 + 自动术语词典
3. 浏览器另开标签页：DeepL 备用翻译
4. 会前：把会议议程贴 ChatGPT 让它准备 5 个可能问题 + 答案模板
5. 会中：Otter 转写让你"看着听"，错过的部分能立刻倒回去
6. 会后：Otter 摘要 + ChatGPT 让它列出 action items

12.12 AI 翻译的"过度礼貌"和"过度啰嗦"陷阱

把中文 prompt 给 AI 翻译，最常见的两个陷阱：

陷阱 1：过度礼貌

# 中文原文
"麻烦你帮我看一下这个 bug。"

# AI 翻译（过度礼貌）
"I would greatly appreciate it if you could kindly take a moment
out of your busy schedule to have a look at this bug at your
earliest convenience."

# 实际工程英文
"Could you take a look at this bug?"

过度礼貌让你显得"不熟悉工程文化"。母语程序员看到 "greatly appreciate" + "kindly" 会皱眉。

陷阱 2：过度啰嗦

# 中文原文
"修复了登录 bug。"

# AI 翻译（过度啰嗦）
"In this commit, we have implemented a fix for the bug that was
affecting the login functionality of the application."

# 实际工程英文
"fix(auth): handle expired token in login"

陷阱 3：错配语体

# 中文是 standup 口语
"昨天搞定了支付回调。"

# AI 翻译（书面语）
"Yesterday, I successfully completed the implementation of the
payment callback functionality."

# 应该是
"Got the payment callback wired up yesterday."

解决方案：精确化 prompt

# Bad prompt
"翻译成英文：<text>"

# Good prompt
"Translate to English. Context: I'm telling my manager during
standup. Tone: casual, terse, like a senior engineer giving a
quick verbal update. No more than 25 words. No 'In this update'
or 'I would like to inform you'.

Text: <text>"

12.13 完整 20 条 Prompt 模板库

把以上和补充的 prompt 整理成 20 条速查表：

12.14 30 天程序员英语训练计划

30 天每天 30 分钟。设定 alarm，把它当跑步打卡一样规律执行。

Week 1：输入 + 命名 / 注释（基础）

Day 1  10min 读 React 最近 10 条 commit + 5min 模仿写 1 条 + 15min 用 prompt 1 让 AI 检查
Day 2  10min 读 Linux kernel 最近 10 条 commit + 改 5 条你过去的 commit
Day 3  10min 读 Stripe API 文档一个 endpoint + 15min 模仿写自己 API 一个 endpoint
Day 4  15min 读 Rust error E0382 + 15min 写 5 条好的错误信息
Day 5  20min 读 PEP 257 + 10min 用 docstring 风格重写自己一个函数
Day 6  20min 读 Conventional Comments 文档 + 10min 给同事 PR 留 5 条 review
Day 7  整周复盘：写 200 字英文反思（不要中翻英，直接英文）

Week 2：写作（PR / Issue / 文档）

Day 8  20min 读 React 最近 5 个 PR 描述 + 10min 模仿写自己一个 PR 描述
Day 9  20min 读 Vue / Vite 最近 5 个 issue + 10min 改写自己一个 issue
Day 10 30min 给一个开源项目提一个 typo / docs PR
Day 11 20min 读一个项目的 README + 10min 改写自己项目的 README 第一句
Day 12 30min 写一篇 500 字英文技术日记（今天调试的问题）
Day 13 20min 用 prompt 14 起草一篇 1500 字博客大纲 + 10min 修改
Day 14 整周复盘 + 完成博客并发到 dev.to / 博客

Week 3：听 + 说

Day 15 30min 看一个 conf 演讲（无字幕过一遍 → 有字幕再看一遍）
Day 16 30min Otter 转写自己 standup 录音 + 找 fillers
Day 17 30min 录自己讲 5 分钟 "tell me about yourself"
Day 18 30min 用 prompt 16 模拟 standup × 3 次
Day 19 30min 听一集 Software Engineering Daily
Day 20 30min shadow 跟读一段 5 分钟英文音频
Day 21 整周复盘：录一段 90 秒 self-intro，对比 day 17 进步

Week 4：面试 + 协作

Day 22 30min 写 STAR 故事 #1（冲突）+ 用 prompt 8 让 AI 评分
Day 23 30min 写 STAR 故事 #2（失败）+ AI 评分
Day 24 30min 写 STAR 故事 #3（领导）+ AI 评分
Day 25 30min 用 prompt 9 做一道 System Design（URL 短链）
Day 26 30min 用 prompt 10 做一道 LeetCode think-aloud
Day 27 30min 写一封 cold email（用 prompt 13）+ 真的发出去
Day 28 30min 在 GitHub 找一个 "good first issue" 并占坑
Day 29 30min 写一份 1-on-1 计划，按照第 8 章的脚本
Day 30 完成 day 28 的 PR，并写一篇英文反思 30 天的成果

12.15 长期英语成长地图

30 天是起步。真正的英语成长是一个 5-10 年的渐进过程：

12.16 工具栈推荐

# AI
- ChatGPT Plus / Claude Pro: 英语写作主要助手
- Cursor: 编码时同步润色注释
- Linear: 用英文写 ticket 强制训练

# 写作
- Grammarly: 实时检查
- Hemingway Editor: 可读性检查
- DeepL Write: 替代 Grammarly 的更"自然"选项

# 阅读
- Pocket: 收藏英文文章
- Reader (readwise.io): 高亮 + 复习
- 语雀 / Notion: 个人英文模板库

# 听力
- Otter.ai: 会议转写
- Listening drill apps: ELSA Speak
- 播客: SE Daily, Changelog, Latent Space

# 词汇
- Anki: 间隔重复你自己的术语卡
- Vocabulary.com: 学术词汇

12.17 最后的话

这本手册到这里就结束了。但程序员英语不是一本书能学完的——它是一项习惯。

真正改变你英语水平的，不是这 12 章的内容，而是你接下来 30 天每天做的事：

• 每条 commit 都用英文写，哪怕是个人项目。
• 每个 PR 描述都按 Why/What/How/Test plan 模板。
• 每周提一个开源 PR，哪怕是改 typo。
• 每月写一篇英文博客，哪怕没人看。
• 每次 standup 都用第 8 章的三段式。
• 每次 review 都用 Conventional Comments 标签。

这些动作单独看都很小，加起来 12 个月——你就会从 "中国程序员" 变成 "能跟全世界程序员一起工作的中国程序员"。

这个差距，比你想象的值钱。

12.18 全书小结

• Ch1-2：建立"程序员英语"的认知 + 命名注释的地基。
• Ch3-5：commit / PR / review 三件套——每天用到的核心。
• Ch6-7：阅读 + 写作的输入输出闭环。
• Ch8：远程会议的口语脚本库。
• Ch9：错误信息和调试英语——被低估的工程美德。
• Ch10：开源协作——长期声誉投资。
• Ch11：技术面试——高强度英语演出。
• Ch12：AI 时代的协作模式 + 30 天训练计划。

祝你成为那个 "代码漂亮、PR 描述清楚、standup 不慌、面试自如、开源圈有名字" 的中国程序员。