Steve Yegge Says the Coding Interview Is Dead. He Might Be Right.

The whiteboard era of reversing binary trees might finally be ending. Steve Yegge — the former Google and Amazon engineer whose blog posts have shaped two decades of industry thinking — just published “The Last Technical Interview,” arguing that AI has quietly killed the hiring process every engineer has spent their career grinding for. It’s a provocative claim. It’s also harder to dismiss than it sounds.

Why the Death Notice Lands Now

For 20 years, the Big Tech interview playbook barely changed. Grind LeetCode. Whiteboard the algorithm. Talk through system design. Pray you don’t blank on dynamic programming. Then 2024 happened. Cursor, Claude Code, and GitHub Copilot graduated from autocomplete novelties into agentic coding tools that solve typical interview problems in seconds.

Yegge’s core argument is brutally simple: the skill being measured is no longer the skill being hired. Memorizing patterns to bang out a correct solution in 30 minutes is a benchmark for a job that doesn’t really exist anymore — not when every senior engineer at every serious company now writes code with an AI pair sitting on their shoulder.

What It Means When the AI Solves the Test

George Hotz recently dropped a video called “how I actually use agentic coding” that pulled in over 140,000 views, and the workflow he demonstrated is genuinely jarring. Hotz barely types code. He describes intent to an agent, reviews the output, redirects, and ships. The keyboard is for prompts and pull request reviews, not for-loops.

The implication is uncomfortable for anyone who’s invested years in interview prep. The premium is shifting from “engineers who can write code” to “engineers who can precisely define what the AI should write.” Asking a candidate to reverse a linked list in 30 minutes tests a muscle that increasingly atrophies on day one of the actual job.

An engineering manager at Meta made a similar point in a recent interview circuit thread on Hacker News: as AI absorbs the repetitive junior-level implementation work, the bar at interviews is sliding from execution to judgment.

So What Should You Actually Measure?

Yegge’s proposed replacement is the interesting part. He thinks the next-generation interview converges on three things.

First, problem definition — can the candidate translate a fuzzy product requirement into a precise technical spec? Second, code review at AI speed — can they spot the subtle bug in a 300-line diff the model just generated with full confidence? Third, systems-level thinking — can they hold the architectural trade-offs the AI can’t see, the ones that span services, teams, and quarters?

Translation: the interview of 2027 looks like the interviewer handing the candidate a laptop with Claude or Cursor already open and saying, “Here’s the problem. Show me how you work.” Not banning the tools. Evaluating fluency with them.

The Market Is Already Moving

There are real counter-arguments. The strongest is that without fundamentals, you can’t catch the quiet bugs the model hallucinates with a straight face. Fair. But the directional shift is hard to deny. Some 2025-vintage startups — including a handful of YC companies — have started explicitly encouraging candidates to use AI tools during interviews. Not tolerating. Encouraging.

Junior hiring is taking the hardest hit, and the math is uncomfortable: the tasks AI absorbs first are precisely the tasks junior engineers were hired to do. Meanwhile, at the staff and principal level, compensation for engineers who can genuinely orchestrate AI agents is going vertical. The middle is hollowing out fast.

The Real Takeaway

Yegge’s framing sounds melodramatic until you sit with it. The technical interview isn’t disappearing. The definition of what a technical interview is measuring is being rewritten in real time.

The formula of grinding 200 LeetCode problems to land at a FAANG is probably already obsolete; we just haven’t admitted it out loud. The new fluencies — pair-programming with an agent, decomposing ambiguous specs, reading generated code with productive skepticism — are what the next decade of engineers will be hired on. If you’re interviewing right now, or running a loop, the honest question isn’t whether to adapt. It’s whether you’re already behind.