If Coding Is Solved, Why Are We Still Coding in Job Interviews?
Oct 28 · 29 min read
A few mid-tier news items dropped today; aside from a few eye-catching mathematical verifications, Google Genie, Artificial Analysis, and FrontierCode all rolled out major version updates to their core products. These have all been AINews headlines before, though they’re mostly annual incremental improvements.
Past podcast guest Felix Rieseberg took today’s top headline, publishing a post highlighting a path forward for junior engineers: “Hunger beats tenure”.
This month, I completely killed traditional coding interviews. As a hiring manager at Anthropic, I no longer require candidates to go through live coding assessments…
…testing you on a simple React challenge is just as pointless as testing everyone on assembly; Claude is now better than humans at both. The success of our most impactful developers here has zero correlation with how well they do on these tests.
Enough said.
His Advice for Junior Engineers
-
Choose business impact over the backlog: Back in 2011, well before the era of AI coding, Patrick McKenzie wrote that “Engineers are hired to create business value, not to write code.”
-
Pick a radical idea: Brainstorm a few ideas that, if successful, would dramatically increase the odds of your team hitting its goals. Be ambitious!
-
Build it over a weekend: I’ve seen countless small teams and solo engineers completely bypass the usual grind of writing PRDs, seeking approvals, and lobbying for designer or PM bandwidth simply by building a working prototype.
-
Focus on one good idea: In my experience, a common mistake junior engineers make is cramming too many ideas into a single prototype, burying 3 brilliant concepts under 20 mediocre ones.
-
Leverage your fresh perspective: If you work on user-facing products, you possess a unique empathy for the user, giving you a much sharper sense of whether the product actually feels good to use.
Well said, Felix!
Get clean, structured web search results in Markdown format—saving roughly 50% on tokens on average compared to JSON. Seamlessly integrate Markdown search results into your AI workflows across 100+ SerpApi APIs!
swyx note: We’re proud to partner with AI-native sponsors you can actually use in your engineering workflows! Check out the token savings from Markdown—slashing token counts on Google data from 361k down to 38k!
AI Twitter Recap
Genie 4: DeepMind’s Real-Time World Model Adds Audio, Physics Simulation, and Gemini Planning
- Launch: Google DeepMind unveiled Genie 4, calling it the “most reali…”