Jensen Huang says the junior developer problem ends in two years. Here’s his math.

Nvidia CEO Jensen Huang is pushing back on the idea that if AI writes most code, software engineers become optional. In a New York Times interview with Ezra Klein at Nvidia’s Santa Clara headquarters, he said people are confusing tasks with purpose. Radiology still needs radiologists, he argued, even if AI handles image scans. Software, in his telling, works the same way.

His line is simple: engineering existed before programming, and it will exist after programming too. Huang said engineers invent products, solve problems, and connect technology to social needs. He pointed to his own early career, when software tools were limited or absent, as proof that the work can survive a change in tooling. He was much less sympathetic to jobs where the task and the role are basically identical, like phone-based customer service. Those, he said, are the ones AI can wipe out.

The sharper part of the exchange was about juniors. Klein asked whether companies still need the same number of new graduates if agents can do so much of the first-year grunt work. Huang’s answer was blunt: “Wait two years.” His math leans on college timing. Since degrees take four years, he said, students who start learning alongside capable agents now should reach the workforce around 2028 with an advantage. He expects that cohort to be “a wave of amazing engineers.”

That confidence comes with a tradeoff, and Huang did not dodge it. He said some “finer intellectual dexterity” will be lost, but better systems thinking will replace it. Klein countered with a study of 26,000 Chinese students that found AI use lifted homework scores by 18% while cutting monthly exam scores by 20% within six months. Huang accepted that some lower-level skill will fade. He just thinks the higher-level payoff is worth it, even if the apprenticeship ladder gets kicked away for a while.

The problem is that coding agents don’t just remove busywork. They also eat the jobs where people learn to spot bad code, trace bugs, write tests, and clean up the mess. Huang’s own framework admits that most of that is task work, which makes it the first thing to disappear. That is the part the industry keeps waving past while talking about productivity, and it is exactly where the talent pipeline starts to leak.