Post by Arseny Kravchenko

Couch prediction: classical ML/DL skills are about to become the Fortran of tech. No hype, but steady demand for greybeards who actually read Hastie, Bishop, and Goodfellow. Here's what's happening. There's a finite population of people who can do ML. The top percentile is building foundation models at top labs - forget about them. The rest of us mortals suddenly found ourselves in demand for the GenAI gold rush. Never mind that tuning gradient boosting and CNNs has little to do with rewriting prompts - hiring managers see "AI" on both sides and connect the dots. So fewer people are left doing actual ML/DL work. But the problems didn't go anywhere. Sure, some stuff got replaced by an API call. Some will get one-shotted by coding agents. But a whole universe of tasks remains - ranking, recommendations, fraud detection, deeply domain-specific problems most of us never think about. Some teams quietly bring so much value through ML optimization that the typical GenAI use case looks like a rounding error on their P&L. And here's the thing: fat AI budgets mean companies can now fund practical, boring ML work with the change left over from their OpenAI contract. Meanwhile the actually tedious part - writing connectors and plumbing - gets easier, not harder. The irony: GenAI hype is quietly turning classical ML into the stable, boring career it never was during its own hype cycle ten years ago. Just nobody's posting about it on LinkedIn.