Anthropic Drops Coding Interviews; Initiative and Prototypes Replace LeetCode as the Differentiator
Hand-written coding interviews mispredict real performance because AI removes the implementation bottleneck, so hiring and career advancement will reward engineers who can identify problems, scope tightly, and prototype with AI instead of relying on tenure or algorithm recall. The essay maps a practical path for juniors to turn a lack of status into an advantage: a tangible prototype over a weekend can replace years of tenure as proof of capability.
Hand-written coding tests no longer predict on-the-job performance now that AI can generate implementations in minutes. Anthropic hiring manager Felix Rieseberg made that call, and the essay extends the argument: when writing code becomes cheap, companies care more about whether you can spot a real problem and turn it into a working result. Junior engineers, previously hamstrung by a lack of tenure and resources, can now use AI to build a clickable prototype over a weekend and use that evidence to win buy-in.
The essay's advice for new engineers is concrete: understand the business goal before touching a PRD, pick one idea with visible impact, build a minimal prototype that answers a single question, and keep the scope tight so the core value stands out. Unfamiliarity becomes an asset, because newcomers see user friction that veterans have normalized. Portfolios should demonstrate taste, empathy, execution, and principled thinking, and should document the problem, why it matters, what was built, how it was verified, and what tradeoffs remain. Coding fundamentals still matter, though: the ability to judge safety, security, and maintainability of AI output is a non-negotiable skill.
The 'hunger beats tenure' idea is really about using AI to generate evidence before asking permission; a two-day prototype can substitute for years of accumulated credibility in a way that was impossible before AI.
Scoping a prototype to one core question is a discipline senior engineers already practice, but AI makes it feasible for juniors to complete in days instead of months, which is why the essay's advice aligns with actual industry workflows.
The essay carefully avoids saying coding fundamentals are obsolete; it shifts the bottleneck from writing code to judging code, which is a more precise and actionable claim than the usual 'AI will replace programmers' narrative.
The thread splits on what still matters when AI writes the code. One view dismisses career advice into development as actively harmful, a 1949-KMT-style trap. Against that, the need to understand principles and read AI-generated code remains a baseline, and the real currency shifts from retrieval-based knowledge to judgment and design. A satirical note rounds it out: the actual differentiator may be self-funding your own AI tools.
Still trying to convince people to do development is pointless, like telling someone to join the Kuomintang in 1949.
Personally, no matter how AI develops, technology still needs to be tinkered with and learned. At least we need to know the principles and how to use it. During development we don't have to code ourselves, we can use AI to code, but at least the code AI writes should be something I can read, or roughly understand, so I'm not unfamiliar with the technology. Whether frontend or backend, I need to be familiar with it. I'm not saying I need to master it, but at least I can recognize it when I see it. Right? Then develop with AI.
The part hit hardest by AI is actually 'retrieval-based knowledge'. My current practice is to delegate tasks like finding tools and looking up solutions to navigation sites and documentation sites, and spend the saved time on judgment and design. The points examined in interviews will likely gradually shift in this direction.
What really sets people apart is whether you're willing to pay for various AI tools yourself, working at your own expense 😁😁😁😁.