Anthropic Manager Abolishes Coding Interviews: AI Makes LeetCode Obsolete

Introduction

In late October 2026, Anthropic engineering manager Felix Rieseberg announced on his personal blog that his team would completely abandon traditional programming interviews — no LeetCode, no algorithm questions, no whiteboard reversal of linked lists. The news ignited debate across the tech community. Some applauded; others scoffed. Regardless of stance, the move reflects a fundamental question: when AI can solve most coding problems in seconds, what exactly are we testing?

This article dissects Felix's original post, analyzes why traditional coding interviews are failing, explores the emerging AI-native engineer competency model, reviews alternative interview approaches being tried across the industry in 2026, and examines the practical impact on job seekers and hiring teams.

1. The Incident: An Anthropic Manager's "Rebellion"

Felix Rieseberg is no outsider. He spent years at Slack before joining Anthropic to lead engineering teams. On October 27, 2026, he published a blog post titled "Hunger Beats Tenure," subtitled "Advice for Junior Engineers." The most striking passage reads:

"I no longer include traditional coding questions in my team's interviews. I don't care if you can produce an optimal dynamic programming solution in 45 minutes. I care about what you've built, how you think, and the judgment you've demonstrated in real projects."

His logic is straightforward: in 2026, a candidate can open Cursor or Claude during an interview and generate a standard LeetCode Hard solution in 30 seconds. The muscle memory cultivated by months of grinding is outperformed by AI a hundredfold. Continuing to test this amounts to evaluating "who is better at pretending AI doesn't exist."

Felix is not the first to think this way, but his identity gives the signal unusual weight — he works at Anthropic, the company behind Claude. When the creators of AI themselves conclude that traditional coding interviews are unsustainable, the signal is strong.

2. Three Structural Defects of Traditional Coding Interviews

Criticism of coding interviews is not new, but the 2026 tooling environment has escalated these issues from "controversial" to "unignorable."

First: What Is Tested Is Severely Disconnected from Daily Work

The vast majority of engineers never hand-write a red-black tree or implement a shortest-path algorithm from scratch. Roughly 80% of work time goes to understanding requirements, reading others' code, designing interfaces, debugging production issues, and making technical decisions. LeetCode tests "implement an algorithm from scratch under time pressure," which has almost no overlap with real engineering scenarios.

Second: AI Tools Have Fundamentally Changed Coding Itself

In 2026, developers using Cursor, Copilot, or Claude Code write code in a radically different way than three years ago. The differentiator between experts and novices is no longer "can you write it" but "can you judge whether AI-generated code is correct," "can you decompose vague requirements into executable AI instructions," and "can you quickly locate the problem when AI makes a mistake." Traditional interviews assess none of these abilities.

Third: Grinding Creates a Massive "False Positive" Market

Many candidates spend hundreds of hours grinding, perform brilliantly in interviews, yet prove weak in actual engineering after hiring. Conversely, engineers with rich project experience and strong judgment are filtered out at the first gate because they lack time to grind. Felix explicitly states he has seen too many "interview kings, project disasters."

3. Reconstructing the Engineer Competency Model for the AI Era

If algorithm questions are dropped, what should be tested? This cannot be improvised; it must start from "what capabilities does an excellent engineer need in 2026?" The following diagram illustrates the shift from the traditional paradigm to the AI-native paradigm:

The core change can be summarized in one sentence:

From "can you write it" to "can you make the right decisions."

Specifically, the three most valuable capabilities in the AI-native era are:

Judgment — Given a vague requirement, can you decide which technical approach to use, which pitfalls to avoid upfront, and when to refactor versus when to ship?

AI Collaboration — Can you efficiently leverage AI tools to accelerate development while also reviewing AI-generated code for correctness and security?

Portfolio — Do you have tangible real-world projects that prove your output in actual engineering environments?

4. Alternatives: What the Industry Is Trying

Felix is not alone in exploring alternatives. In 2026, a growing number of companies are experimenting with new interview processes. The diagram below summarizes the mainstream patterns:

Four concrete approaches:

Portfolio Review

Instead of posing problems, candidates bring their own projects to discuss. Interviewers probe: "Why was this architecture chosen? What pitfalls did you encounter? How would you redesign if traffic grew 10x?" The focus is not on code aesthetics but on whether the candidate genuinely reasoned through engineering decisions.

AI-Paired Coding

Candidates receive a moderately complex task and are encouraged to use Claude, Cursor, or similar tools. The evaluation centers on how they decompose the problem, instruct the AI, review the generated code, and handle AI errors. This may be the most forward-looking format of 2026 because it tests exactly what engineers do daily.

Real-World Scenario Simulation

Candidates are given a buggy production code snippet to debug, or a pull request to review. These tasks closely mirror daily work and are difficult to game through rote memorization.

Paid Trial Work

Some companies — especially small-to-mid-sized teams — offer 1–3 day paid trials where candidates complete a small feature in the real codebase. This enables mutual evaluation: the company observes actual output, and the candidate assesses code quality and team culture.

5. Practical Impact on Job Seekers and Hiring Teams

For Job Seekers

The signal is clear: do not put all eggs in the grinding basket. LeetCode is not useless, but its ROI is plummeting. In 2026, higher-return investments include:

Building one or two deep personal projects or open-source contributions where you can articulate the full technical decision chain in an interview.

Mastering at least one AI coding tool — not merely "knowing how to use it" but efficiently driving it and reviewing its output.

Participating in real engineering — even fixing bugs, writing docs, or doing code reviews for open-source projects beats solving the 500th LeetCode problem.

For Hiring Teams

Reforming the interview process is not easy. The greatest strength of traditional coding interviews is standardization — every candidate solves the same problem under a unified rubric, enabling easy cross-candidate comparison. Portfolio reviews and scenario simulations introduce more subjectivity and demand stronger engineering judgment from interviewers themselves. Felix acknowledges this raises the bar for interviewers, but argues that is precisely the point: if your interviewers lack engineering judgment, the hires won't either.

Another practical issue: not every candidate has a showcase portfolio. Fresh graduates and career switchers may have limited project experience. For them, AI-paired coding interviews are a valuable supplement — they assess potential and thinking style rather than accumulated credentials.

6. Closing Thoughts

Felix Rieseberg's abolition of coding interviews is not a claim that "algorithms don't matter." Algorithms are the bedrock of computer science. But what interviews test and how they test it must evolve with the era's tools and work practices.

In 2016, testing LeetCode made sense — engineers truly needed to hand-write large amounts of code from scratch, and raw speed plus algorithmic fundamentals directly affected output. Continuing that approach in 2026 is like requiring manual stick-shift parallel parking in the age of autonomous driving — the skill isn't useless, but it's no longer the core differentiator between good and bad drivers.

Felix ends his post with a line that resonates:

"Hunger beats tenure."

Hunger — passion for technology, desire to build great things — outweighs seniority. Traditional coding interviews measure neither hunger nor true engineering ability; they only measure willingness to spend three months memorizing problems.

It's time to change how we test.