I interviewed for a product designer role at Anthropic last month. I've done design interviews at Meta, Figma, and a few startups, and this one was distinct enough that I wanted to write it up.
context: mid-senior product designer, 8 YOE. the role was focused on the Claude.ai consumer product.
the process:
portfolio review: 60 minutes. unlike some places where this is mostly a presentation, Anthropic's felt more like a Socratic seminar. I'd describe a decision, they'd push back: why not this other direction? what did you learn that surprised you? what would you do differently? they dug hard into one case study in particular, the one I thought was most straightforward. pick the work you can defend most thoroughly, not the shiniest outcome.
design exercise: take-home, 3-4 days. mine was: design an onboarding experience for a new Claude feature that helps users accomplish a goal they haven't clearly articulated yet. which is a legitimately hard problem. I spent real time on it. they're not looking for pixel-perfect Figma mocks, they want your thinking process documented. show how you got to the decision, not just the decision.
design critique: they showed me a UI from the Claude product and asked what I'd change and why. this isn't a trick question. the point is to see how you articulate a design problem and prioritize. I saw a couple things that seemed inconsistent with a coherent information architecture and said so. they seemed interested in that framing.
behavioral: same mission/values depth you'll hear about in all the Anthropic write-ups. for designers they also asked about how I think about the ethics of UI design, specifically around AI. things like: how do you think about anthropomorphization in how you write interface copy? when is it irresponsible to make an AI feel too human?
comp was in the 180-230k TC range. equity is the usual pre-IPO variable.
one last thing: they're genuinely thoughtful about design at Anthropic. the questions they asked me reflected a team that has argued about these problems seriously. it wasn't a rubber stamp process.