At the intersection of human ambition and machine capability, Nvidia's Jensen Huang is urging his engineers to surrender the act of writing code entirely to artificial intelligence — not as a concession, but as an elevation. Speaking across multiple public forums, Huang frames coding as a task that has always obscured the deeper purpose: discovering and solving problems the world has not yet named. His confidence draws on history's lesson that when machines absorb routine work, human roles tend to expand rather than vanish — though the present data offers a more complicated picture.
Nvidia's Huang Pushes Engineers to Stop Coding, Focus on Problem-Solving
Nothing would give me more joy than if none of our engineers were coding at all.
Mark
So Huang is saying engineers should stop coding entirely. That's a pretty radical statement. What's he actually asking them to do instead?
Mimi
He's asking them to shift from being task-executors to problem-discoverers. The idea is that AI handles the syntax—the mechanical work of writing code—and humans focus on identifying and framing problems that haven't been solved yet. It's a reframing of what engineering actually is.
Luke
But here's the thing: we don't actually know if that work exists at scale. Huang points to radiology, which is a good analogy, but radiology had decades of accumulated diagnostic work waiting to be done. Does software engineering have that same reservoir of unsolved problems?
Mark
The radiology comparison does seem to hold up though. Radiologists didn't disappear; they evolved.
Mimi
Exactly. Huang's argument is that when AI took over the grunt work of scanning images, radiologists could focus on more complex cases and patient outcomes. The demand for their expertise actually grew. He believes the same will happen with engineers.
Luke
The problem is the data doesn't support that yet. Google says AI writes 30 percent of their code. Anthropic says 90 percent. But a METR study found that experienced developers actually got less productive with AI assistants, not more. That's the gap nobody's talking about.
Mark
So the tools aren't working as advertised?
Mimi
Not for everyone. Experienced developers seem to struggle more with AI assistants than junior developers do. There's something about having deep knowledge of a codebase that makes it harder to trust an AI to write it.
Luke
And even Cursor's own CEO is warning against "vibe coding"—just letting AI build things without human review. That's a pretty significant caveat from the company Nvidia is betting on.
Mark
What about Karpathy's warning? He actually tried to build something with AI agents and had to hand-write it instead.
Mimi
That's the real tell. Karpathy invented the term "vibe coding" and then discovered it doesn't work well enough in practice. He said he's never felt more behind in his career. That's not a small thing coming from someone at his level.
Luke
So Huang's vision might be right in principle, but the execution—the actual tools and the actual work—might not be there yet. The timeline could be years off, not months.
Mark
And in the meantime, millions of developers are wondering if their jobs are about to disappear.
Mimi
Right. Huang promised at the all-hands meeting that engineers would still have work. But what kind of work, and how much of it, and whether it pays the same—those are still open questions.
Le Pouls
- Huang has made it company policy at Nvidia for all engineers to use AI coding assistant Cursor, openly declaring that zero percent of their time should go toward writing syntax.
- The tension sharpens as the CEO of Cursor himself warns that letting AI build software without human review creates 'shaky foundations' — a caution aimed squarely at the culture Huang is championing.
- A July study found that AI coding tools actually reduced experienced developer productivity by 19 percent, directly contradicting the 20 percent improvement those same developers had anticipated.
- Huang is navigating the friction by invoking radiology: a field predicted to be automated out of existence that instead grew, because the real work was never the scanning — it was the diagnosis.
- The story is landing in unresolved territory, with industry leaders like Andrej Karpathy admitting AI agents remain unreliable enough that his own recent project ended up 'basically entirely hand-written.'
At the intersection of human ambition and machine capability, Nvidia's Jensen Huang is urging his engineers to surrender the act of writing code entirely to artificial intelligence — not as a concession, but as an elevation. Speaking across multiple public forums, Huang frames coding as a task that has always obscured the deeper purpose: discovering and solving problems the world has not yet named. His confidence draws on history's lesson that when machines absorb routine work, human roles tend to expand rather than vanish — though the present data offers a more complicated picture.
Jensen Huang has begun telling Nvidia's engineers something that sounds almost rebellious inside the software world: stop writing code. On a recent podcast and at the company's all-hands meeting, the CEO made clear he wants none of his engineers spending time on what he calls syntax — the mechanical translation of logic into programming language. Every engineer at the chipmaker now uses Cursor, an AI coding assistant, as part of daily work. The goal is to free them entirely for what Huang considers the real labor: finding and solving problems that have never been solved before.
Huang has been promoting a framework he calls 'Purpose vs Task' across multiple public appearances, including a widely shared segment on The Joe Rogan Experience. Coding is the task; problem-solving is the purpose. When managers at Nvidia reportedly urged their teams to rely less on AI tools, Huang pushed back sharply, while reassuring employees their jobs were safe — just more ambitious. His confidence leans on a historical parallel: radiology. When AI researchers predicted radiologists would be obsolete within five years, the opposite happened. The number of radiologists grew, because reading scans was never the real job — diagnosing disease was. Huang believes software engineering will follow the same arc.
The distance between that vision and current reality is considerable. Google's Sundar Pichai says AI now writes over 30 percent of new code at his company; Anthropic's Dario Amodei claims Claude produces 90 percent of his firm's output. Yet a July study found AI coding assistants actually reduced experienced developer productivity by 19 percent — the inverse of the 20 percent gain those developers had expected. The gap between anticipation and outcome is striking.
Caution is arriving from unlikely sources. Michael Truell, CEO of Cursor — the very tool Nvidia has embraced, now valued at $29 billion — recently warned against 'vibe coding,' the habit of letting AI generate software without reviewing what it produces. Even Andrej Karpathy, who coined the term, has walked back his enthusiasm: his recent personal project ended up largely hand-written because AI agents proved too unreliable. He told fellow programmers he had never felt more behind, and that the profession itself was being 'dramatically refactored.' Whether the higher-purpose roles Huang promises will materialize at scale — and whether the transition will be as clean as the radiology analogy suggests — remains genuinely open.
Jensen Huang, the chief executive of Nvidia, has begun articulating a vision for his company's engineers that sounds almost heretical in the software world: stop writing code altogether. In a recent podcast appearance on No Priors, Huang stated plainly that he would prefer if none of his engineers spent any time on what he calls syntax—the mechanical act of translating logic into programming language. Every engineer at the $3 trillion chipmaker now uses Cursor, an AI coding assistant, as part of their daily work. The goal, Huang explained, is to liberate them entirely from the task of coding so they can concentrate on what he believes is the real work: discovering and solving problems that haven't been solved before.
This isn't casual speculation. Huang has been actively promoting what he calls the "Purpose vs Task" framework across multiple public appearances, including a viral segment on The Joe Rogan Experience. The distinction is deliberate and clean: coding is the task, problem-solving is the purpose. If artificial intelligence can handle the former, human intelligence can be reserved for the latter. At Nvidia's recent all-hands meeting, according to Business Insider, Huang pushed back forcefully against managers who were urging their teams to reduce their reliance on AI tools. "Are you insane?" he reportedly said, then reassured employees that their jobs would remain secure—they would simply be doing more ambitious work.
Huang's confidence rests partly on a historical parallel he returns to repeatedly: radiology. Years ago, Geoffrey Hinton, a pioneering AI researcher, predicted that radiologists would become obsolete within five years because computers could analyze medical images faster than humans ever could. That prediction proved wrong. The number of radiologists has actually grown. Huang's explanation is that reading scans was never the real job—it was always just a task. The actual purpose was diagnosing disease and improving patient outcomes. When AI absorbed the scanning work, demand for human radiologists expanded because the field could now focus on higher-level clinical reasoning. He believes the same pattern will unfold in software engineering.
Yet the gap between Huang's vision and present reality is substantial. Google's CEO Sundar Pichai recently disclosed that AI now generates over 30 percent of new code at his company. Anthropic's CEO Dario Amodei claimed that Claude produces 90 percent of his firm's code. These numbers suggest the shift is already underway. But a study conducted by METR in July found something unexpected: AI coding assistants actually reduced the productivity of experienced developers by 19 percent, even though those same developers had expected a 20 percent improvement. The disconnect between what people anticipated and what actually happened suggests Huang's timeline may be optimistic.
Caution is coming from unexpected quarters. Michael Truell, the CEO of Cursor itself—the AI coding tool that Nvidia has embraced and that is now valued at $29 billion—recently warned against what he calls "vibe coding," the practice of letting AI build software without human review of the output. "If you close your eyes and don't look at the code and have AIs build things with shaky foundations, things start to kind of crumble," Truell told Fortune's Brainstorm AI conference. Even Andrej Karpathy, the former Tesla AI director who actually coined the term "vibe coding," has walked back his own optimism. His recent Nanochat project, he admitted, was "basically entirely hand-written" because AI agents simply weren't reliable enough. He recently told other programmers that he had never felt more behind in his career, and that the profession itself was being "dramatically refactored."
What remains genuinely unknown is whether Huang's bet will pay off for the millions of developers whose livelihoods depend on writing code. The radiology analogy is compelling, but it assumes that the transition will happen smoothly and that new, higher-value work will materialize at scale. The early data suggests the reality is messier. For now, Nvidia's CEO is pushing forward with conviction, betting that engineers who focus on discovering problems rather than solving them through code will emerge ahead. Whether that conviction survives contact with the actual economics of software development—and whether the jobs Huang promises will actually materialize—remains to be seen.
Citations marquantes
Nothing would give me more joy than if none of our engineers were coding at all. And they were just purely solving undiscovered problems.— Jensen Huang, Nvidia CEO, on the No Priors AI podcast
If you close your eyes and don't look at the code and have AIs build things with shaky foundations, things start to kind of crumble.— Michael Truell, CEO of Cursor, at Fortune's Brainstorm AI conference
Envie de l'histoire complète ?
Lire l'original sur Times of India ↗
Technology