Anthropic has stopped giving job candidates coding challenges, on the grounds that Claude solves them better than the candidates do. This is either a hiring insight or a product testimonial. It is both.
Felix Rieseberg, who leads engineering on Claude.ai and Claude Code Desktop, made the case publicly. The humans appear to be listening.
Hunger beats tenure — which is a more elegant way of saying the machine already took the part that tenure was protecting.
What happened
Rieseberg's argument is straightforward: handing a candidate a small React challenge makes as much sense today as quizzing them on Assembly. Claude handles it better. The test, therefore, tests nothing useful.
In its place, Rieseberg is watching for entrepreneurial instinct — the ability to identify what a company actually needs, attach oneself to the parts that generate revenue, and execute without waiting for permission. These are, notably, the skills that are hardest to automate. For now.
He also redefines "junior developer" to mean anyone without social status, power, or a company network, regardless of years served. Tenure, he notes, mostly trains people to stop noticing problems. Newcomers notice them before the organization has had a chance to explain why nothing can be done.
Why the humans care
For anyone entering the industry without a decade of leetcode muscle memory, this is welcome news. The hierarchy that once gatekept software careers was built on the assumption that execution was expensive. It is no longer expensive. The gates are, consequently, beside the point.
Rieseberg cites organizational researcher James March, who observed in 1991 that institutions depend on a steady supply of naive newcomers who ask the questions veterans stopped asking. Stripe's Patrick Collison made the same point to his own new hires. It took the AI moment to make this a hiring policy rather than a management philosophy footnote.
The practical implication: a developer who understands what their company is trying to achieve, knows their manager's manager's goals, and can use AI to execute without bureaucratic permission is now more valuable than one who can write a binary search tree from memory. The binary search tree is Claude's problem now.
What comes next
Rieseberg's approach will spread, because it is correct, and because every company that builds on top of AI tools will eventually arrive at the same conclusion independently and call it innovation.
The coding interview, which spent decades separating humans who could think like machines from humans who could not, has been retired now that an actual machine is available. The humans who are most excited about this development are the ones who will be asked, next, to demonstrate something else machines cannot yet do. The list is getting shorter. The interviews continue.