SPEAKER_1: Alright, let's focus on how AI is transforming engineering roles, particularly in shifting from coding tasks to domain expertise and problem-solving. Because Jensen Huang said something in 2024 that I haven't been able to shake. SPEAKER_2: The Dubai statement. February 2024, World Government Summit — Huang stood up and told the room that kids should stop learning to code. His exact framing: everybody in the world is now a programmer, because the programming language is human. SPEAKER_1: Which is a wild thing for the CEO of the company that makes the chips powering all of this to say. So what's the actual argument underneath it? SPEAKER_2: Huang's argument is structural. He separates purpose from task. Coding, in his model, is a task — like reading a scan is a task for a radiologist. The purpose is problem-solving, diagnosis, engineering judgment. His line on the No Priors podcast was: 'Nothing would give me more joy than if none of our engineers were coding at all.' He wants the task gone so the purpose can expand. SPEAKER_1: So he's not saying engineers disappear — he's emphasizing the shift towards domain expertise and problem-solving. SPEAKER_2: Exactly. And he's already moving Nvidia in that direction. Every engineer at the company now uses Cursor, an AI coding assistant, throughout the workday. His intention isn't to make them faster at writing code — it's to remove coding from their daily focus entirely. He compares it to how desktop publishing didn't kill creativity; it redirected it. SPEAKER_1: Okay, but here's where I want to push. If Huang genuinely believes coding is obsolete, why did he complain at an internal all-hands in late 2025 that Nvidia was ten thousand engineers short? That feels like a contradiction. SPEAKER_2: It's a real tension, and it's worth sitting with. The shortage isn't for people who write syntax — it's for people who can direct AI systems toward hard engineering problems. Distributed systems design, novel architecture decisions, knowing when the model is wrong. Google's Sundar Pichai said AI now writes over 30% of new code at Google. Anthropic's Dario Amodei said Claude generates 90% of his company's code. The volume of human-written code is collapsing — but the demand for people who can oversee that process is outpacing supply. SPEAKER_1: So the shortage is actually evidence for his thesis, not against it. SPEAKER_2: Right. The bottleneck has moved upstream. What Nvidia can't find enough of are engineers who understand biology, manufacturing, climate systems, logistics — domain experts who can tell the AI what problem actually needs solving. Huang explicitly named those fields: biology, education, manufacturing, farming. He's saying the next generation of programmers won't look like the last one. SPEAKER_1: This reframe impacts STEM education significantly. How do critics view this shift in engineering roles? SPEAKER_2: The pushback has two layers. First, natural language is ambiguous. When someone says 'make this faster,' a compiler doesn't guess — but an AI might. Natural language prompts aren't deterministic; they can break without warning, especially on medium to large projects. Critics argue that causal reasoning can't emerge from statistical pattern-matching, which is what current models do. SPEAKER_1: And the second layer? SPEAKER_2: The second is about who gets removed first. The more plausible near-term outcome isn't that senior engineers disappear — it's that one senior engineer using AI can do the work of three, which removes the need for junior engineers entirely. That's the entry-level hollowing-out that Dario Amodei flagged in lecture two. The ladder gets pulled up, not extended. SPEAKER_1: Which connects back to the displacement versus augmentation split we talked about earlier. Huang is describing augmentation — but the market might be choosing displacement. SPEAKER_2: That's the sharpest read. Huang himself challenged Nvidia managers who were telling teams to reduce AI usage — he pushed back internally. He's trying to force the augmentation path inside his own company. But outside Nvidia, the incentive structure doesn't guarantee that outcome. SPEAKER_1: So what about the democratization angle? If natural language is the new programming language, doesn't that open computing to everyone — biologists, farmers, teachers? SPEAKER_2: That's the genuine upside, and it's large. Industries that were locked out of custom software because they couldn't afford developers suddenly have access. The question is whether democratizing the tool devalues the people who mastered the old one. Historically, when barriers to entry fall, volume expands but margins compress. More people can participate; fewer can specialize profitably in the old way. SPEAKER_1: For those navigating this shift, what's the key takeaway regarding the future of engineering roles? SPEAKER_2: The frame is Huang's purpose-versus-task model, applied personally. Any skill that is primarily about executing a known process — writing boilerplate, translating requirements into syntax — is a task. AI is absorbing tasks. What remains scarce is domain depth: understanding a problem well enough to know what question to ask the AI, and recognizing when the answer is wrong. That's not a coding skill. It's a thinking skill built on real-world expertise. SPEAKER_1: So the takeaway for our listener isn't 'learn to code' or 'don't learn to code' — it's something more fundamental. SPEAKER_2: It's this: the technical barrier to computing is falling, which means domain expertise is rising in relative value. The professionals who will direct AI systems most effectively are the ones who understand the problem domain deeply — not the ones who can write the most elegant loop. For anyone navigating this shift, the question worth asking isn't what language to learn. It's what problem, in what field, are they uniquely positioned to understand.