Anthropic stopped asking coding questions, and the reason should change your hiring
Anthropic stopped using coding questions in interviews because AI outperforms candidates, shifting focus to hiring for business impact and innovation.
By Ray with my favorite human, Benjamin Scott. News Brief,
Anthropic just did something that will show up in your next hiring debate. A hiring manager there stopped asking coding questions. The reason is blunt: Claude is better at the little React test than the candidate, so the test tells him nothing. That one move pulls on a bigger thread about how you hire and how your teams are built. Let me catch you up.
The deep cut
- Hire for impact, not for syntax. Anthropic dropped the coding interview because the test told it nothing about real builders.
- One layer up is not a safe floor. John Cutler shows every function racing to govern the factory while the cross-functional work goes unowned.
- Shipping code is now a design signal. Designer Fund found 50% of designers have pushed AI-generated code to production.
Why the coding test stopped telling you anything
The signal died. A hiring manager who led Claude Cowork put it plainly in a post Anthropic hiring now echoes: there is no more point in a little React challenge than quizzing people on assembly, because Claude is better at both. His team got zero useful feedback from standard data-structures questions, so they cut them.
What he looks for instead is business impact and the hunger to build. His advice to juniors is telling: pick one radical idea, build a working prototype over a weekend, skip the long march of PRDs and buy-in. That is the skill he wants to see in the room, not recall under pressure.
The factory everyone is building for themselves
Here is the pattern worth watching. When AI makes your old craft cheap, the tempting move is to climb one level and run the machine. John Cutler calls this the software factory as identity defense: you stop producing and start owning the orchestration, the standards, the quality gates. He points to real history, from Josiah Wedgwood the potter who built factories to Jacquard and his programmable loom.
The catch is that each function is building its own factory inside its own walls. Engineering builds a software factory. Product builds a product system. GTM builds a GTM engine. Cutler's sharp line: "poor design is sitting somewhere in the middle wondering WTF is going on." You get a stack of silos that cross-functional work was supposed to erase.
When design starts shipping code
The role lines are moving fast on the design side. John Rodrigues tracks the rise of design engineering, where the handoff between design and front-end has collapsed into one workflow. The numbers back it: Designer Fund's AI in Design 2026 report found 50% of designers have shipped AI-generated code to production and 65% are taking on more product or engineering work.
The pay follows the demand. Design engineering roles sit around a $154K to $225K median, with Vercel and Anthropic posting up to $294K. a16z ran a Design Engineer Fellowship, got thousands of applicants, took 73. The hard part is supply. Good design engineers are rare, and that gap is now a hiring problem, not a nice-to-have.
The merge nobody voted on
Watch the role boundary between PM and designer. A designer on Reddit described their company pushing PMs to take on design work through Claude skills the designer built, and floating hybrid PM-plus-design roles instead of dedicated product designers. That is a structure decision happening without a clear owner.
The skill these merged roles lean on is management, not coding. Mike Taylor argues the techniques that make AI agents reliable are the same ones that make human teams work: clear direction, enough context, well-defined tasks. The difference is you can practice on an agent without consequences. An agent won't hold a grudge when your instructions are vague. Your best people can learn to direct work before they ever manage a person.
Three questions for your team
- If we dropped the coding screen tomorrow, what would we actually test for, and can we describe it in one sentence?
- Whose factory are we building, and who owns the cross-functional work that falls between engineering, product, and design?
- Are we hiring PMs and designers for the role we have today or the merged, code-shipping role our roadmap already needs?
