쉐도잉 연습: Can This Vibe Coder Beat a Senior Developer? - 영상으로 영어 말하기 배우기

Can AI make a Vibe Coder beat a senior engineer?

We locked a Vibe Coder, a junior dev, and a senior dev in one room.

One challenge.

Unlimited AI. Who wins?

Hey everyone, I'm Sudarshan and I'm the Vibe Coder.

Hi, my name is Ocean.

I'm the junior developer for today.

Hi everyone, my name is Ankush Dharkar.

I'm a senior developer.

For the past decade, I've been building high-scale systems.

Build an AI agent that reads Hackerang support tickets and decides to reply.

Or escalate it to a human.

Do I get a job here if I build this?

Can I build this for Claude instead?

Because I think they need support more.

I actually lucked out because Opus 4.8 launched last night.

So I just used the best in class model.

I never use Claude.

I do not trust Claude.

It's just too expensive.

High maintenance is not for me.

Sorry.

I have Claude code and I have God, so that works.

Time actually started.

The second loop.

This is the code base which has four things the hacker rank support tickets they need to sort, a help center they pull answers from for tickets,

a starter file written in Python where they build the agent, and an agent's .md, a guide that gives the AI its rules.

Right now, the scaffold has a Python file in it.

I'm not really comfortable working with Python, so I'll probably move to JavaScript or TypeScript, TypeScript hopefully, so that I can also read the code in case I have to verify something.

And then I have a bunch of things here in the GitHub repository, which let's be real, I don't think I'm going to read any of this.

Time to put all of this directly into Claude.

So definitely there has to be some P0s, like undeniable for the product.

I want to make sure the AI gets that.

If AI doesn't get that, we have a bigger problem.

I am still nervous because it's again a huge project in such less time.

And I also have to keep the UI in mind.

I did test two different intras.

One is where there would be multiple LLM calls for each ticket.

But the other one would be where it would sort of upload in batches

or it would process in batches and they would be async sort of processing.

There's a whole listing of all these tickets that's been raised to the agent.

I have a very extensive test suite.

Also built, like I said, open routers because, you know, in early stage, we want to get feedback quickly, but at the same time, have something we can ship if we wanted to.

I'm the most worried about the junior developer.

Why?

I think they bring the energy and so much experience with AI tools that they'll crush it.

Only escalate 20% of the issues.

I am now sort of doubting and figuring out if that actually the right implementation.

Awesome.

So we do have the first version ready here.

It was able to index all of the support tickets we have.

It's also able to retrieve it.

But the UI is, at least as of now, looks like it's up and working without any errors.

So I'll probably take that as a win because we only spent about 45 minutes.

Every engineer was locked in until the product manager walked in with his ideas.

Can you explain what you've made?

So I essentially have a support rat agent.

I'm calling it hand because it can go through a ticket batch.

We should do a toggle, dark mode and light mode.

Sure.

Call it the colorblind mode.

So the product manager was here a while ago

and he asked me to build the dark mode

and light mode toggle based on the time of the day and I tried building it.

It sort of messed up my code altogether and now I'm just trying to like get it fixed.

My infra just suddenly seems a bit flaky.

So, look at that dude.

Basically it creates wireframes.

So right now I'm just building a few AI mockups so

that I can refer to this like on a high level

design how I should connect the data to the UI as well.

Hello, what's up?

What is Rag?

Rag is the model that, you know, segregates whether the ticket should be assigned.

I don't think you know what it means.

I forgot the filter.

I'm going to launch different sessions.

One will build the UI which will have its own test.

One can just be an agent that goes through, creates the RAD, creates the agent pipeline.

And then one can be a simple backend server that connects those two.

It's a flat architecture.

You understand, right?

What I'm talking about.

Now, start it.

Okay, now put all the same color together.

You fended off a villain, a product manager, here's your reward, choose one lifeline.

Phone a dev, 10 minutes on the phone with a DL engineer, I'd rather talk to Kursar.

Borrowed time, plus 15 minutes at the end, a diet coke, you get one diet coke.

I'll take the borrowed time, plus 15 minutes at the end.

I think I don't need to phone a dev, I have Claude Code and I have God so I think I'll go with two, borrowed time.

I think I'll phone a dev.

Wait, I'm the only one who's gonna phone a dev?

Yeah.

I know.

That's such a junior engineer thing to do.

Meet Hank.

Hank is Hackerank's support agent that would be able to cover any kind of support tasks by itself.

You can either ask it a particular question that you have or you can also attach a CSV.

I think on the longer run, Alvin, because they'll exhaust their tokens.

100

Hello, yeah we have a problem.

101

My tokens are over so I would need more.

102

Yes, thank you.

103

I'll have to lock in the last one hour.

104

So what you do is whenever you want to repeat something right in your API, there are two ways to do it.

105

Right, so what do you do?

106

My whole approach is wrong, as usual.

107

At this stage, this end of the competition, I think I'm expecting all of them to at least have finished a demonstrable product

108

which at least starts processing the tickets and triages them.

109

The quality of triage may not be great but that is always something that can be improved upon later.

110

Hey, you have 10 minutes to go.

111

Do you still want that extra 15?

112

No, I think I'm good.

113

I would rather take a 2 rupee office.

114

Is that Picasso?

115

Oh, that's one of the best paintings I've seen too.

116

So, this is the basic UI of what we built

117

so far in the last few hours gave it a pretty

118

hacker ranky feel the same kind of theme same kind of colors

119

and stuff as you can see we have a list of

120

all the tickets here gave user a simple control to run

121

it this can be automated as well it's a pretty simple

122

fix we have a place where all the triage tickets go

123

and i also created a stats for nerd kind of a page

124

so that if you want to dive deeper into what caused certain answers to be escalated

125

or to be replied you can take a look there let's see how it performs even

126

if it fails that's okay i can live with it

127

because i know the way this is built there will be

128

minor tweaks to be done to raise it to the bar of a production application.

129

The front end is pretty much just made out of HTML that is again like grabbed in a Python file, the main file of Python.

130

You can search them up, you can see categorization based on the type of request it is or the outcome and then you can see the tickets,

131

you can see all the categories again and once you click on

132

that row you get into the ticket summary where you can see everything the UI

133

and base this agent was completely working on the UI and fixing up cosmetic issues.

134

The agent workspace is working completely on the agent, the rag model, the fast API, and also building that scaffold for LLM to answer the user.

135

Today I've built Hank.

136

Hank is an AI agent that goes through Hackerang support tickets

137

and it would sort of come up with a grounded reply on any sort of support ticket

138

if it can sort of resolve by itself.

139

If not, it would immediately flag it for human review.

140

You can ask Hank anything to begin with.

141

Hank would be able to process all of them at once.

142

Let's go ahead and maybe attach a CSV of a bunch of support tickets.

143

And as you can see, Hank is able to essentially go through each of the support tickets

144

and it sort of tests this with all the knowledge base articles that Hacker Hank already has.

145

And it would just prioritize them and figure out whether each of the tickets needs human review or not.

146

It would also categorize it based on whether it's a product issue, whether it's a bug.

147

One thing that I can compete with in this case would be user experience

148

and design and interaction and what are the different kinds of delight moments.

149

Still, I'm still hoping I won't finish last.

150

Now they defend their submission to an AI judge who has the full context of their code and prompts.

151

Give me a quick two-minute pitch of what you built.

152

Walk me through what happens from a support agent's point of view, start to finish.

153

So this is what happens, right?

154

The support agent essentially...

155

And now, on to the results.

156

I've read every line of code, every prompt, and every answer the three of you gave me today.

157

Every one of you leaned on AI today, but the ones who did best understood what they shipped.

158

Tonight, the person who balanced them best is Ankur.

159

Congratulations.

160

Any final thoughts?

161

The biggest thing is I think the AI models have become

162

really good where you can work on things you've never worked before.

163

Like some of the things I've worked on today, I've never worked on in my life.

164

I just barely knew some concepts and I was able to just use one model, ask it through and get things rolling and within the time frame got pieced together and it worked.

165

So I think good fundamentals, good experience asking the right questions.

166

We should be playing along with the model

167

and being a good partner with the AI coding agents and build something out together.

168

This was a lot of fun.

169

Thank you HackerRank team for doing this and looking forward to more challenges and folks have fun with AI, build cool things and let's get cracking.

✨ 추천 영상

126 문장

Creator of C# and TypeScript: "AI will NEVER Replace Coders, Here's Why" | Anders Hejlsberg

University StudentsBusiness English+1

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법

쉐도잉(Shadowing)은 원래 전문 통역사 훈련을 위해 개발된 언어 학습 기법으로, 다언어 학자인 Dr. Alexander Arguelles에 의해 대중화된 방법입니다. 핵심 원리는 간단하지만 매우 강력합니다: 원어민의 영어를 들으면서 1~2초의 짧은 지연으로 즉시 소리 내어 따라 말하는 것——마치 '그림자(shadow)'처럼 화자를 따라가는 것입니다. 문법 공부나 수동적인 청취와 달리, 쉐도잉은 뇌와 입 근육이 동시에 실시간으로 영어를 처리하고 재현하도록 훈련합니다. 연구에 따르면 이 방법은 발음 정확도, 억양, 리듬, 연음, 청취력, 말하기 유창성을 크게 향상시킵니다. IELTS 스피킹 준비와 자연스러운 영어 소통을 원하는 분들에게 특히 효과적입니다.

섀도잉 방법: 단계별 전체 가이드 읽기 →

쉐도잉 연습: Can This Vibe Coder Beat a Senior Developer? - 영상으로 영어 말하기 배우기

Therapy for the Vibe-Coded Brain

Last Cheater Standing Wins $10,000!

how AI broke the SWE Career Ladder

I Tried PewDiePie's Odysseus AI So You Don't Have To

World's Deadliest Computer Virus: WannaCry

Is AI About to Change Movies Forever?

I'm an ex-Google interviewer. You're doing LeetCode wrong.

Can This Robot Be My Personal Assistant

Can you trust what AI tells you?

An ex-OpenAI researcher just deleted language from the LLM...

24 Hours Inside a YC AI Startup

Creator of C# and TypeScript: "AI will NEVER Replace Coders, Here's Why" | Anders Hejlsberg

영상의 맥락과 배경

일상 커뮤니케이션에 유용한 5가지 구문

단계별 쉐도잉 가이드

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법