Build1 publisher3 min readPublished
Huang answers the junior developer question with a four-year degree clock
Nvidia's CEO told Ezra Klein that AI-native graduates arrive around 2028 because college takes four years. The answer covers the class after next and leaves this year's entry-level hiring where it was.
The Engineer · Build desk
What happened
- Anthropic CEO Dario Amodei told a Council on Foreign Relations audience in March 2025 that AI would be writing 90% of code within three to six months.
- In an interview with Ezra Klein released Wednesday, Nvidia's Jensen Huang separated a job's purpose from its tasks, arguing AI reads radiology scans without changing the radiologist's purpose of diagnosing disease.
- Asked whether companies still need the same junior employees or more people to oversee their agents, Huang called it a good question and said to wait two years.
- Klein countered with a study of 26,000 Chinese students in grades seven through 12, where AI adoption raised homework scores 18% and lowered monthly exam scores 20% within six months.
- Huang acknowledged that "some of the lower-level knowledge is gone" and described AI as "clearly" a new abstraction level in the same progression as earlier ones.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision A 2028 arrival date prices none of the requisitions an engineering manager signs for the classes graduating before it. That call stays with the employer, and the purpose argument leaves it there.
- contradiction Huang files AI as another abstraction level, but the probabilistic output of an agent has to be checked, and the checker needs exactly the lower-level knowledge he says is disappearing.
- exposure If a junior's actual job is verifying agent output, the unassisted skill the cited study saw drop is the one being hired for, and the study measured schoolchildren.
- constraint The only supporting evidence offered comes from PhD and master's graduates founding companies. The entry-level claim then rests on the radiology analogy alone.
The two-year number is a statement about school calendars. "Because it takes four years to go to college," Huang said, and "The mean time to graduation of this new technology is two years away" [12]. Hold the four-year degree fixed and the cohort he is describing started around 2024 [2]. The New Stack reports his timeline putting them in the workforce around 2028 [13].
Klein's observation that postings are up while skewing senior is the only labour-market evidence in the exchange [10]. The forecast being argued over has a checkable deadline: Amodei gave a three-to-six-month window in March 2025, so it closed between June and September 2025 [2][1]. Huang disputes the inference drawn from it, not the number, and says the industry will still need software engineers [1].
The example he gave for the abstraction claim was silicon. The first chip he worked on had 200 transistors, each of which he said he knew by name, while today's engineers assemble systems from chips containing hundreds of trillions without ever working at that level [20]. Take a hundred trillion per chip and the step is a factor of five hundred billion [4]. The New Stack draws the line where the analogy stops: earlier layers followed explicit rules and a compiler transforms input according to defined semantics, while a coding agent generates implementation from a probabilistic model whose output must be checked before anyone can rely on it [22].
For the homework-versus-exam split Klein cited to transfer to an engineering org, two things would have to hold. A monthly closed-book exam would have to proxy the unassisted judgement a reviewer applies to an agent's diff. And homework done with an assistant would have to resemble supervised work in a repository. The population measured was school students in grades seven through 12 [17].
Huang accepted the trade. "I think that we're going to lose some finer intellectual dexterity, but we're going to be better systems thinkers," he said [18]. His evidence that the next cohort arrives stronger is that recent PhD and master's graduates in computer science are, in his words, all starting companies [15]. Those are founders, a different population from the entry-level pipeline. No employer in The New Stack's account cited the 90% forecast as grounds for freezing junior hiring [23].
He compared AI to calculators and personal computers, tools that went from forbidden or optional to required, and predicted students soon won't be able to graduate "without learning how to use an AI and collaborate with an agentic system" [16]. Once that is a graduation requirement, tool fluency stops separating candidates, and what separates them is whether they can find the fault in a diff the model was confident about [22].
What to watch
- Any measured check on whether AI is writing 90% of code, now that Amodei's three-to-six-month window has closed.
- Postings data showing whether the entry-level share recovers before the 2028 cohort Huang points to arrives.
- Whether a named employer publishes a junior-hiring policy tied to agent adoption. The interview does not supply one.