Andrej Karpathy is just as anxious about AI jobs as we are

Seeing someone like Andrej Karpathy—a guy who literally helped build the foundations of the current AI era—admit he's unsure about the long-term job outlook is a massive reality check. Usually, it's the juniors or the mid-level devs panicking while the "visionaries" tell us everything is fine. When the people inside the room are shrugging, it means the shift is deeper than just "learning a new tool."

The core of the discussion isn't actually about doom, but about the Jevons paradox. For those who aren't econ nerds, the paradox basically says that when a resource becomes more efficient (cheaper), we don't actually use less of it—we end up demanding way more of it. In this context, as AI makes writing code "cheaper" and faster, the total global demand for software and digital products is going to explode.

The catch? The demand isn't for the type of engineer who was hired in 2019.

If your entire value proposition is "I can translate a Jira ticket into a functioning React component," you're in the danger zone because that's rule-based work. AI eats rule-based work for breakfast. The real value is shifting toward "insider judgment"—the ability to navigate the messy, non-linear, human parts of a project that a model can't see.

Think about it like this: an AI can write a perfect contract or a flawless piece of code, but it can't tell you that the vendor you're hiring is untrustworthy because of a weird interaction you had with them in a hallway five years ago. That's the "human layer" that survives the compression.

If you're looking for a practical tutorial on how to survive this, stop focusing on just syntax and start focusing on the AI workflow. The engineers who thrive won't be the ones fighting the LLM, but the ones standing close enough to the mechanism to redirect it. We're moving from being "writers" of code to "editors" and "architects" of systems.

To actually pivot your career toward this, you need a deep dive into how LLM agents actually operate. Instead of just prompting, you should be looking at how to orchestrate these tools to solve business problems that aren't clearly defined. The "rule-based" engineer follows instructions; the "judgment-based" engineer decides what the instructions should even be.

It's a weird feeling knowing that even the OpenAI crowd is feeling the heat, but it's also liberating. It means the goal isn't to "beat" the AI at coding—which is impossible—but to lean into the parts of the job that are fundamentally human. Judgment, intuition, and strategic navigation are the only things that aren't being commoditized right now.