Career advice in the age of AI

There's a quiet shift happening in how people think about career strategy, and it's not about which programming language to learn next. The advice making rounds boils down to something simpler and harder: focus on work that can't be neatly graded within a model's training cycle. If a skill can be reduced to a benchmark, someone is already training a model to beat you at it, probably faster than you'd expect.

That means the tasks with clear right answers, the ones with tidy rubrics and quick feedback loops, are exactly the ones getting automated first. Writing boilerplate code, summarizing documents, drafting standard emails. These are gradeable, so they're trainable. The advice isn't to avoid them entirely, but to stop building an identity around being good at them.

The more interesting shift is about problem-finding versus problem-solving. Most career advice for the last few decades assumed problems arrive pre-packaged: here's a bug, fix it; here's a spec, build it. But models are getting scary good at the solving part. What they're not good at is walking into a messy, ambiguous situation and figuring out which problem is even worth solving. That's a human judgment call, shaped by context, taste, and lived experience that doesn't fit neatly into a training dataset.

There's also a positioning argument buried in here, and it's the one people tend to skip. Being early to notice an opportunity matters more than being technically excellent at executing a well-known one. That requires proximity to where things are changing, not just competence at a fixed task. So the practical takeaway isn't "learn to prompt better" or "study machine learning." It's closer to: go stand where the interesting, ill-defined problems are surfacing, and get comfortable being the one who names them before anyone else does.