A former OpenAI researcher just dropped the most uncomfortable career advice of the AI era.
And it's aimed squarely at anyone under 35.
Meet Phil Chen.
Ex-OpenAI. Ex-DeepMind. Ex-Scale AI. Ex-Helm AI.
Now building his own agent-native startup.
He's watched companies scale from 15 people to 100,000+.
And he says the game has fundamentally changed.
"AI models get better at anything you can write a loss function for. And school is mostly loss functions."
Read that again.
Well-defined problems. Known answers. Clean grading.
That's exactly what AI eats for breakfast.
👉 So the valuable work of the next decade is everything that can't be graded.
At Phil's company, nobody writes code by hand.
So Leetcode? Useless.
System design rounds? Also useless.
Instead, candidates get dropped into a messy environment and judged on:
The skill isn't coding anymore.
It's problem selection and resource allocation.
Before Scale AI, Phil had quant offers with way bigger guaranteed cash.
He said no.
That single decision unlocked DeepMind. Then OpenAI. Then a founder network worth more than any signing bonus.
In 2023, he turned down early Anthropic (~50 people).
He turned down early Cursor (just 2 non-founder employees).
Both are now generational companies.
His point isn't regret. It's this:
Capital is easy. Time, relationships, and reputation are the actually scarce assets.
Phil doesn't buy the doom.
He doesn't think ASI replaces knowledge workers.
Because humans still hold two superpowers machines can't grade:
Choosing what's worth doing.
And deciding where to point the compute.
The next trillion-dollar careers won't be built by the best solvers.
They'll be built by the best problem-finders.
That's all for now!
