There’s a version of the coding interview where you invert a binary tree on a whiteboard while someone watches. Claude passes that test. So does GPT. So does every frontier model shipped in the last two years. At one Anthropic team, the test is now gone from hiring entirely.
Felix Rieseberg, an engineering manager at Anthropic who led Claude Code Desktop, removed the coding interview from his team’s process. His reasoning is blunt: Claude outperforms people on the short coding tests the interview depended on, which means the test no longer measures anything worth measuring. A screen that a machine aces isn’t separating strong engineers from weak ones. It’s separating humans from a benchmark.
What he screens for instead is drive. He’s watched newer engineers “fly past” colleagues who are, in his estimation, better coders. His summary of the pattern is three words: hunger beats tenure.
The uncomfortable context
This would be a standard hiring hot take if it came from anywhere else. It doesn’t. Anthropic co-founder Jack Clark said earlier this year that AI makes the value of junior engineers “a bit more dubious,” and most of the industry discourse has drifted that way ever since. Headlines about dead entry-level jobs. Ghost internships. Seniors-only hiring funnels.
Rieseberg is arguing the opposite from inside the same company. When implementation gets cheap, he’s saying, the scarce skills stop being implementation. They become taste, judgment about what’s worth building, and the nerve to actually ship the thing.
I find his version more convincing, if only because it comes with a mechanism. “Juniors are doomed” is a vibe. “Execution isn’t expensive anymore” is a claim you can build a career strategy on top of.
The weekend prototype playbook
His advice to junior engineers is specific enough to act on this week. Three moves:
Pick one bold idea. A real one, with an actual chance of embarrassing you. Portfolio polish and tutorial rebuilds don’t count.
Tie it to a goal your leadership already cares about. A prototype nobody asked for is a hobby. A prototype aimed at an objective your manager is already accountable for is a promotion case.
Ship it as a weekend prototype. Not a quarter-long project with a steering committee. A weekend. The bar for “shipped” has collapsed, and that cuts both ways: it’s never been easier to build something real, and it’s never been more obvious who isn’t building.
The proof point is sitting on your desktop. Claude Code’s desktop app was prototyped by two newer employees over a single weekend. It went public a little over a week later. No genius required. That’s just what happens when execution cost approaches zero and someone has the appetite to run with an idea.
What this does to hiring everywhere else
Most companies haven’t caught up. Interview loops still burn hours on take-home tests that a chatbot completes in about a minute, then wonder why the signal quality is low. The tests are worse than outdated. They’re inverted: they filter for exactly the skill, fast and correct small-scope coding, that depreciated fastest.
The harder problem is what replaces them. Rieseberg’s answer, drive, is real but expensive to evaluate. You can grade a binary tree in two minutes. Measuring whether someone has the judgment to pick a bold idea and the persistence to ship it takes conversations, references, and ideally a trail of things they’ve built. Hiring gets slower and more human, not more automated. There’s some irony in that.
There’s also a fairness wrinkle. Coding tests, for all their flaws, were at least uniform. “Hunger” is easier to perceive in people who resemble the people already in the room. Any team dropping structured evaluation owes itself a hard look at what subjective screens do to who gets through.
What to do now
If you’re early in your engineering career, the playbook above is the whole post. Pick the idea, aim it at something your leadership already wants, build it this weekend. Your GitHub doesn’t need more tutorial clones. It needs one artifact that makes a manager lean forward.
If you run a team, audit your interview loop for anything a frontier model can complete unassisted. Anything that passes that bar measures nothing. Either redesign it around reviewing a real change the candidate made to a real codebase, or follow Rieseberg and spend the time probing for drive instead. Rote tests are theater now. The teams that figure out what replaces them first get the hungry engineers everyone else is filtering out with binary trees.
