跟讀練習: ai killed software engineering - 通過影片學習英語口語

選擇練習模式

每種模式都會開啟全屏練習室。

230 句

如果句子過短或過長，請點選 Edit 進行調整。

Hey, I know you read the title, you probably think, wow, another guy on YouTube with the AI psychosis.

You're right, wait, no, you're wrong.

I've had the psychosis before and luckily I've overcame it.

This includes all kinds of writing with AI, vibe coding, creating videos.

And, you know, I spend still $3,000 a month on tokens at work.

And I've been a software engineer for five years now.

So I kind of have a view of what it looks like before AI and after AI.

So that's kind of some credentials of why you should still watch the video.

But I wanted to talk about one thing, and that's how AI has killed the joy of software engineering.

Let's get her going.

So a day before and after AI hasn't really changed too many things besides one thing, and that is the amount of code you're reviewing.

And I'll explain why.

So before you spent most of your time coding and working on the implementation yourself, and let's say it was like four out of the eight hours of the day,

you would have a good idea of what you're already implementing.

You would do a quick review of the code if you're submitting it to teammates so they can review.

And then once you do that, it's up to your teammates to then review and approve the change.

And then you get into production.

Now you're spending your time just talking to agents, going through the implementation, hopefully putting in good business context of why you're making a change

and then it's off to the agent to determine what to implement you might review

that and then it's it makes all the changes for you

but you're pretty much hands off between that initial step and

when it finishes and you know during

that time it's hard to just work on one thing right

because an agent might work for 15 to 30 minutes

so you can basically do nothing for that time

or you can try and fill

that in with you know another agent working on something else

which leads into something we'll talk about more at the end of the video.

But code review is still very important in the process because agents are good.

I would say like 80 to 85% of the time, but there's still that 20 to 15% of the work that needs to be adjusted

just to follow best practices or fix bugs that were missed.

And, you know, it's important you do this because you don't want these defects to land up in the user's hands.

So you try and catch these things before you ship this code to production.

But the issue is now

that you're further away from the actual implementation of what the

agent's doing It takes a ton more time to get up to speed on what changes it made

And then go in and actually review that

that code like you should be doing So

if you're actually being a good boy

and you review All of the code it creates which you should be doing because you know,

it's not perfect Then you're spending the majority of your day reviewing this agent output But it's also kind of annoying,

too, because you're looking for just these minor things that's wrong with the agent, like you would be doing if you were reviewing another person's change they were trying to make.

But except now you're just doing that way more.

So it's probably like you're doing double the amount of code review because you're reviewing your own code a lot more now.

And then now you still have to review teammates code.

Code review, I thought, and still is, is one of the worst parts of the job.

It's very tedious and it's still very important to get a good understanding of how your systems work,

but the joy of software engineering was around writing the code, not reviewing the code.

You don't get that same dopamine hit

when you finish reviewing the code with an AI agent compared to when you implemented that feature yourself.

Now, this is just a basic overview of what I think has changed, but there's a lot more smaller stuff that I will touch on that has also led to the joy decreasing.

The next thing I want to talk about is that I've never felt more tired after finishing working than I do now.

There's just an insane amount of drain on your brain after a full day of work

because I'm essentially running multiple agents at a time.

And these agents could be across multiple repositories.

So a repository is just where you put code, but the pieces, the chunks of code could be on totally separate parts of the business.

So you have to keep switching back and forth between these agents and the context that you're working in is totally different.

And there's been a lot of studies that show context switching is not good for productivity.

And I have some thoughts on that.

I feel like we are getting net gains from agents, But I don't think there's been any good studies that have came out and said, yes, we are getting efficiency gains.

But kind of a tangent, just flipping back and forth between these agents just takes a lot out of you

because you have to remember what you're working on and why.

And that's one of the key differences before and after, because before you would just be focused on one feature.

You would get it done and then you would basically ship it off.

Now you're I mean, you don't have to be, but most people are working on multiple things at the same time.

So there's just a ton of context switching.

and now you're personally context switching a lot

but now imagine this your teammates are shipping more and more features

and you have to code review all this stuff

so just imagine all of that context switching you have going on now

because there's so many various things that you're working on

and what i've noticed now is

that people don't even review their code they'll have the agent

implement the code based on what they initially said they'll give it a quick glance be like okay yeah

that looks good then they'll create you know the the change request

and they'll leave it up to the person reviewing the code to actually try

and find stuff

or the person reviewing your merge request is just like yeah it looks good to me approves it

and then you just ship it it's honestly insane but

it just takes so much time to go through all this code

that you know people have to get stuff done themselves i'm

not saying we shouldn't review code like i still review all of the code

that comes through it's just like insane

because you have to be locked in for a full day

to get through all this kind of stuff you need to

get to i also want to touch on another thing kind of related

so as you know you care less about the code you

review less code you review less your teammates code you're going to lose understanding of your systems

and then you're basically in the ai psychosis where it's up to the ai to define implementation

and architecture and all that kind of stuff

and you're just kind of backseating

and letting it do whatever it's just going to be a terrible model because you're going to lose all your domain knowledge.

And at some point, you're going to be completely useless.

So you need to stay in the driver's seat and actually figure out what's going on, have a good understanding of your system so you can design them properly for the future.

I feel like I would be doing the industry a disservice if I didn't bring this up.

And that is AI slop.

And I'm not even talking about coding because, you know, that's an issue too with people vibe coding.

You know, within companies, obviously you can vibe code a random project

and it's totally fine because you're not shipping this to users.

100

So the quality doesn't have to be that great.

101

But I'm just talking about people, you know, shitting out documents about features we should implement

102

that was all written with AI with not even like a thought behind

103

that or people sending messages to you or in a Slack channel that have just been written by an LLM.

104

So they just send this extremely long blob of text.

105

And I'm like, bro, I'm not reading any of that shit.

106

It's just so annoying.

107

and I don't know.

108

What I was saying at the beginning, I did have this psychosis where I thought, okay, you know, it would be cool

109

if I could hook up Slack to an agent

110

and then it could send messages to people and maybe they could have an agent that talks to my agent.

111

And I'm like, bro, at this point, what are we even doing?

112

Like how would their agent have more knowledge than what my agent has?

113

Because all of that context should already be available.

114

And I was like, you know, this ain't it.

115

And then one more thing back on the code changes, people should still be writing their commits and their descriptions for their changes

116

because that actually forces them to have a good understanding of what's being changed

117

and then people still read the description of changes so you don't want someone reading just a huge blob of text, like a huge change log of, you know, what changes were made.

118

They don't care about that.

119

They care about why the changes were made.

120

So you don't want to just put this huge burden on

121

people to just figure out why the changes were made looking at the code.

122

That should be in the description.

123

Another thing that has reduced the joy of the job is just the executives at companies.

124

They hear about all these things at other companies, you know, they might write a blog or a YouTube video or something like that about how they did something with AI.

125

And then every executive thinks they need to do the same thing.

126

When in reality, most of these executives have no clue anything about AI.

127

And they're literally just being sheep, trying to figure out, you know, they're seeing all these efficiency gains or whatever that other companies are reporting with AI.

128

And they're like, oh, wow, we need to do this.

129

So they're like, you know, everything needs to be done with AI, crazy psychosis coming down from leadership and

130

that just gets reflected down the entire chain

131

and then it gets down to the devs

132

and the devs are you know ai is pretty solid i think everyone will admit

133

that but like using it for everything is just annoying

134

and like the executives don't understand that there's still some other elements

135

that shouldn't be used for ai but i feel like that's another thing

136

that just causes a lot of burnout is just the executives not understanding

137

and give and giving devs like the free will to do whatever they want.

138

Like they understand that these tools are good so they'll use them for that

139

but you don't need to like mandate stuff and force people to use whatever.

140

And obviously this will vary a lot by company

141

but it's I think it's pretty common from what I've seen online

142

that a lot of the companies are doing this now

143

which is unfortunate and it's kind of wild because they're spending

144

so much on tokens and I don't think anyone's actually like seen the efficiency gains yet, or at least reported that.

145

No studies or no companies have come out and said that.

146

And I think the next wave is going to be a huge token budget across all companies.

147

It's going to be wild.

148

I know Uber already came out with that, but that's going to be the next thing because I'm spending $3,000 a month on tokens, which I would say is a lot.

149

But if everyone at your company is doing that, that's a lot of money to be spending on that, especially when you can't prove out the efficiency gains.

150

So I feel like every company will have to take a second look at that.

151

And all the mandates coming down, all that stuff, they'll probably just end up with token budgets.

152

And that'll be interesting to see how that plays out.

153

Another thing related to executives is just the amount of pressure they've put on people now with AI.

154

You know, they hear all this stuff in the industry about how it's so good and you get all these gains.

155

so then they're like okay we either don't need as many people

156

or we just we're gonna ship way more features

157

so then anytime you push back on like a project saying

158

we need more time on this they're basically like oh don't

159

you have ai like you can easily get this done

160

and it's the same thing when you're like down a man on your team

161

and you're trying to get another person they're like oh can't you just make up for

162

that person's workload with ai it's like come on insane But yeah, back to my previous point about people not reviewing code,

163

it's kind of hard when you have this pressure down from leadership about, oh, we want to launch these 10 features this quarter, however they're saying it.

164

And then actually taking the time to understand what is changing, because you have to keep iterating on stuff and getting these features out to hit these deadlines.

165

So you don't have as much time to actually take a deep look at the code.

166

and eventually people will get destroyed from this because they're just offloading everything to the AI so they're just kind of cooked.

167

Another interesting thing related to that point is

168

what I've noticed is there's only so much to build for like an industry or domain or company

169

and ideas are expensive now, like good ideas.

170

So it's almost like yes you can implement so much faster now

171

and i don't think anyone really disagrees with that

172

but the end of the day you actually need to have good product vision

173

and have a good idea that you're implementing

174

like so many companies are just trying to layer ai onto their entire business

175

but it's like okay first off the trust for ai is going way down right now

176

and then does ai actually make sense for your use case

177

you know it might make sense for like 10 of the use cases

178

but they're throwing ai literally onto everything and it's like users don't want

179

that so why are we implementing a feature

180

that people will not use you know back to the psychosis

181

all good i feel like we have to touch on the software engineering community too

182

and honestly it's just a clown show first off i think it's

183

because anyone that has a normal take on social media is not going to get any views

184

so So you kind of have to be on like one of the sides and be polar.

185

But I don't know, man.

186

There's the side where it's like, okay, AI is trash.

187

It's not useful at all.

188

And there's a side of like, oh, everyone's cooked.

189

AI is so good.

190

Like AGI is coming.

191

And I don't know, man.

192

I don't know how they just can't realize that, you know, these tools are useful, but they're not like taking anyone's job yet.

193

It's just kind of unfortunate.

194

I feel like that leads to devs just being annoyed at AI because anywhere you go, it's like one side or the other.

195

you go on social media

196

and it's people posting like oh my god this is the

197

most insane workflow like you're never gonna need to code anymore

198

all this kind of stuff like oh here's a new model

199

that dropped like your job's cooked it's just I don't know it's not very good like I wish people

200

that had real takes actually you know got the visibility for

201

that because I don't know it just also leads to the draining of being a dev hearing this all day

202

and then you got the executives seeing this shit on social media

203

and they're like oh why are we not doing x we're

204

doing y like we should do x you know what i

205

mean it's just not it i will say one thing

206

that has remained the joy is

207

if at work you can find a product you like working on

208

because if you like what you're building

209

and you can build towards what users want i feel like

210

it's that's the satisfaction i see in ai now since you can build so much faster.

211

The satisfaction is no longer in writing the code.

212

It's just satisfying users and building, you know, a product that users love, building new features that they'll actually use.

213

And, you know, it's getting ridiculously tough to build anything useful because there's just so much stuff out there.

214

And what I mean by that is like building, you know, a SaaS tool that will scale, which is why I've been, you know, focusing on personal tools lately.

215

There's some stuff that I would want.

216

I can tailor that to how I would use it.

217

And I don't need to adapt my, you know, thinking to some platform someone created or some tool someone created,

218

which I might make a video on next explaining some of the stuff I've been working on.

219

But yeah, I feel like that's where most of the joy is going to come from.

220

And the unfortunate thing is I feel like most people work for companies

221

and products that they don't really care about and not blaming anyone, but most devs I feel like just work to collect a paycheck, which is totally fine.

222

I wouldn't say I'm not that either, but yeah, it's just kind of unfortunate.

223

But let me know what you guys think.

224

Are you still having joy in software engineering?

225

I think I kind of am, but a lot of it has been decaying.

226

And with all that said, it's still one of the best jobs out there that pays in saying, I personally get to work from home.

227

I don't know, solid work hours, so you can't complain.

228

I'm just one of the things I've noticed as AI has progressed over the last couple of years.

229

And lastly, I would advise getting familiar with these power tools, just in case your job gets taken by AI.

230

Peace.

下載應用

流利口語所需的一切

AI 發音評分逐句打分

音標練習掌握每個音

詞彙學習建立你的詞庫

詞彙遊戲邊玩邊學

掃碼

Download on theApp Store Get it onGoogle Play

FOR YOU

本課的詞彙與口語要點

這節 B2 級別的口語課以影片“ai killed software engineering”為素材。影片中反覆出現的詞有：code, agent, review。這段影片共有 230 個句子、3160 個單詞可供跟讀。講話部分時長為 15:52。說話人語速較快，每分鐘約 199 個詞，會出現較多連讀和弱讀。 88% 的單詞屬於英語最常用的 3,000 詞，詞彙容易理解。

影片中的重點詞彙

影片中最難的 15 個單詞，附發音和釋義：

單詞	發音	釋義
implement 動詞	/ˈɪmpləmənt/	實施 /實施, 執行 /執行
psychosis 名詞	/saɪˈkoʊsɪs/	精神錯亂 /精神錯亂, 精神症
token 名詞	/ˈtoʊkən/	代幣 /代幣, 籌碼 /籌碼
teammate 名詞	/ˈtiːmˌmeɪt/	隊友 /隊友
annoy 動詞	/əˈnoɪ̯/	攪擾 /攪擾, 煩擾 /煩擾
unfortunate 形容詞	/ʌnˈfɔː.t͡ʃə.nɪt/	不幸
vibe 名詞	/vaɪb/	氛圍 /氛圍
repository 名詞	/ɹɪˈpɒz.ɪ.t(ə.)ɹɪ/	倉庫 /倉庫
mandate 名詞	/ˈmændeɪ̯t/	授權 /授權, 權利 /權利
approve 動詞	/əˈpɹuːv/	批准, 贊成 /贊成
satisfaction 名詞	/sætɪsˈfækʃən/	滿意 /滿意, 滿足 /滿足
burnout 名詞	/ˈbɜː(ɹ)naʊt/	倦怠
credential 名詞	/kɹɪˈdɛnʃəl/	證書
dopamine 名詞	/ˈdoʊ.pəˌmin/	多巴胺
tangent 名詞	/ˈtæn.d͡ʒənt/	切線 /切線

影片中出現的短語動詞

單詞	發音	釋義
figure out 動詞		弄清楚
come out 動詞	/ˌkʌm ˈaʊt/	出, 出來 /出來
go through 動詞		通過 /通過
come down 動詞		下來 /下來, 降
get through 動詞	/ɡɛt ˈθɹu/	度過 /度過
get up to 動詞		幹 /幹, 做
hands off		勿動手！ /勿動手！, 不許碰！ /不許碰！

值得反覆跟讀的句子

影片中簡短而完整的句子，可以直接用在日常對話裡：

You're right, wait, no, you're wrong.
There's some stuff that I would want.
Are you still having joy in software engineering?

影片中的語法

說話人最常用的結構，並附上影片中的原話：

結構	影片中的用法
現在完成進行時 have/has been + -ing — 以前開始、現在仍在繼續的動作	I've been working · has been decaying
現在完成時 have/has + 過去分詞 — 過去發生但與現在仍有關聯的事	I've had · I've been · has killed
被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的	be adjusted · were missed · be locked
定語從句 who / which + 從句 — 補充說明人或事物	scale, which is · paycheck, which is

需要注意的發音

說話人用了 116 次縮略和弱讀形式，例如 you're, don't, they're。請按聽到的簡短形式來說。

“sh” 和 “zh” 音: satisfaction /sætɪsˈfækʃən/, credential /kɹɪˈdɛnʃəl/, sheep /ˈʃiːp/
長單詞——注意重音位置: unfortunate /ʌnˈfɔː.t͡ʃə.nɪt/, repository /ɹɪˈpɒz.ɪ.t(ə.)ɹɪ/, satisfaction /sætɪsˈfækʃən/, ridiculously /ɹɪˈdɪk.jə.ləs.li/, visibility /ˌvɪz.əˈbɪl.ɪ.ti/

如何用這段影片練習

先完整聽一遍影片，不要開口，記下不認識的單詞。
先用 0.75 倍速逐句跟讀，熟練之後再回到正常速度。
錄下自己的聲音並與原聲對比，特別注意 implement, psychosis, token 這類單詞。

什麼是跟讀法？

跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧，最初開發用於專業口譯員的培訓，並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大：您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同，跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性，語調，節奏，連讀，聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。

影子跟讀法: 閱讀完整分步指南 →

跟讀練習: ai killed software engineering - 通過影片學習英語口語

選擇練習模式

下載應用

推薦影片

Anthropic just proved AI isn't getting better

Spec-Driven Development: AI Assisted Coding Explained

How I'd Learn n8n if I had to Start Over in 2026

8 Tech Choices to Lock In Before Agentmaxxing

AI Coding Is a Dead End

Virtual love - How dangerous are AI relationships? | DW Documentary

The Only SaaS Moats Left in 2026

15 Minutes to Fix Your AI Dev Workflow with OpenCode

Ultimate Claude Guide: How to Use Claude AI for Beginners in 2026

Is ML Engineering Dead? (AI vs ML Engineer)

Jack Dorsey’s New Dessa AI Voice Sounds Scary Human

How to Become an AI Product Manager in 2026 | Ex-Google, Microsoft