AI News Daily한국어

The generation that starts as team leads: KAIST's Prof. Minhyuk Sung says "'GPT did it' is no excuse"

Nov 20, 2026Original 프로페썰(KAIST 성민혁 교수) · The Daily Californian · Howdy(해커랭크 비벡 라비산카르 CEO) ↗

In 3 lines

  • With AI doing team members' work, new grads start as team leads and need hands-on sense to direct it
  • F rate in a Berkeley intro course hit 35.3%, under 10% the two prior years; AI cheating and overreliance blamed
  • HackerRank's CEO says new-grad hiring is the highest in four years; fundamentals plus orchestration is rare
The generation that starts as team leads: KAIST's Prof. Minhyuk Sung says "'GPT did it' is no excuse"

"From now on, you start as team leads." That is what Prof. Minhyuk Sung of KAIST's School of AI Computing tells students in their twenties. On the Korean YouTube channel Professul on November 19, he spent 18 minutes on entry-level workers in an era when AI does the team members' work, how AI is changing the classroom, and a usage gap that tracks how much students can pay. We paired his examples with reporting from UC Berkeley's student newspaper and hiring data shared by HackerRank's CEO.

"You start as team leads"

Sung believes that as lower-level work shifts to AI, everyone now has to work one level up. People used to learn the job as team members before becoming managers, but today's twenty-somethings are a 'special generation' who must direct AI team members from day one. To act as a lead, you need to predict what AI does well and badly and what it will bring back, so the skills asked of newcomers have actually grown. He added that schools and companies should create ways to gain team-member experience quickly.

"Never mind, I'll ask my AI agent"

Sung shared a 'scary story' he heard recently. A manager asked a team member how some code was going; the team member, away on a business trip, suggested meeting later, and the manager replied, "Never mind, I'll just ask my AI agent." Instead of asking the person about their work, the manager asked the AI. Sung worried that the trial and error and communication people used to learn by working through things together are fading.

Flashy assignments reveal 'AI slop' within a minute

The classroom has changed too. Students bring in plenty of impressive-looking work, Sung said, but a minute of looking or talking makes it obvious when AI produced it carelessly. What matters far more now is showing you can explain, adapt and manage what you made. Host Lee Kwang-yong called this trend 'the outsourcing of cognition.'

35% F's in a Berkeley intro course, after two years under 10%

The UC Berkeley case mentioned in the talk was reported in June by The Daily Californian. In spring 2026, 35.3% of CS 10 students and 10.6% of CS 61A students received F's; in spring 2024 and 2025, neither course exceeded 10%. Dan Garcia, who taught both, attributed it mainly to a sharp rise in cheating with large language models and said nearly 30 CS 10 students were caught cheating on take-home exams. Sung said professors and students alike agree the only fix is to put away laptops and phones and work by hand in the classroom.

The Daily Californian · The Daily Californian ↗

A 20-year-old interview question: low level or abstraction?

Sung recalled a question from his master's admission interview that now looks prophetic. In the future, a senior professor said, only two kinds of people will remain: those who understand technology deeply at the low level and those who work broadly at a high level of abstraction. Which would he be? He answered "low level" and says he is still on that path. His students are splitting the same way, and he argued schools should support both.

Push orchestration, and students "just use it"

Sung pointed to 'harness engineering' as today's keyword: splitting work among an agent that writes code, one that reviews it and one that runs it to find problems, then making them interact, much like a lead building a team. He encourages students to use AI aggressively, but after stressing it, some began churning out output without checking whether it is right or meaningful. The ability to verify that the AI's work matches your intent is becoming more important, he said.

300,000 won of a 700,000 won allowance goes to AI

Students worry AI use is creating a class divide. One raised a hand in class to say that 'if you have money, you get good grades now' and asked how the school was preparing. A 20-year-old freshman Sung met spends 300,000 won of a 700,000 won monthly allowance on AI first, studying with eight windows open on screen. Sung said schools must give students a chance to use a decent level of AI without paying out of pocket.

Counterpoint: "Not hiring new grads is overdone"

Other data complicates the story of a closed entry door. Vivek Ravisankar, CEO of coding-assessment platform HackerRank, said in a conversation released November 18 that "we should stop hiring new grads is overdone," and that by HackerRank's usage, new-grad hiring this year was the highest of the past four years. Companies that want to become AI fluent need AI-native people, he said. Still, someone who can only vibe code without fundamentals is in a tricky zone, while those who combine orchestration with software engineering fundamentals are a "very precious" asset.

Howdy · YouTube ↗

"GPT did it" is no excuse, and the first generation that can do both

When students answer a problem with "GPT did it," Sung tells them not to talk about GPT as if it were a person. A person did the work, and taking responsibility and verifying it was done right is where humans come in. This generation has to master fundamentals and AI at once, which is hard, but no earlier generation could build both abilities together. It may be a generation with opportunities its predecessors never had.

Comments 0

Select a sentence in the story (long-press on phones) to comment on it.

Original프로페썰(KAIST 성민혁 교수) · The Daily Californian · Howdy(해커랭크 비벡 라비산카르 CEO)↗

More insights