Felix Rieseberg, an engineering manager at Anthropic, published a post on October 27, 2026 saying he retired the traditional coding interview for his teams. Candidates no longer solve a small programming task in the room, because Claude solves those tasks better than they do. What he reads instead is whether an applicant built something unprompted, and what that thing says about their judgment.
What was actually released
An essay, on the author’s personal site, filed under management. No product, no pricing, no corporate process change. Rieseberg writes from inside Anthropic, where his bio says he has led engineering for Claude.ai, Cowork and Claude Code Desktop.
The argument has three parts. The test is gone on his teams: he says a small React challenge makes as little sense today as quizzing a candidate on assembly, because Claude is better at both. The people who impress him are often the ones with the least tenure. And if execution is cheap, the scarce skill is picking what to execute.
What the essay actually claims
- A coding challenge no longer separates strong candidates from weak ones on his teams.
- Recent engineers are overtaking better coders who have been around longer.
- The replacement signal is a working prototype built unprompted, read for taste and empathy.
He also redefines the word at the center of the post. Junior does not mean young or newly graduated. He means engineers without social capital, status, or power, which includes people who have been at a company for years.
The practical part is a five step sequence, and it is the most reusable thing in the piece.
| Step he gives | What it replaces | Evidence he offers |
|---|---|---|
| Attach to business impact | Working the backlog in order | A 2011 line from Patrick McKenzie: engineers are hired to create business value, not to program things |
| Pick a radical idea | Fixes nobody has to approve | Ideas that would dramatically improve the odds the group hits its goals |
| Build it over a weekend | Specs, buy-in, lobbying designers | Prototypes that skip the requirements document |
| Keep it to one idea | Three good ideas buried among twenty bad ones | A prototype that also needs a new language, a backend and mobile apps fails |
| Use the fresh perspective | Deferring to whoever came first | The Notion offline mode he was told was impossible |
What the claim rests on
Almost nothing in the post is measured. No count of candidates seen before and after, no comparison between test scores and later performance, no data showing that people hired on prototypes do better. The evidence is anecdote, and the anecdotes are good: at Notion he was told an offline mode was impossible, asked the founder, and found the resistance was assumed rather than real.
The strongest claim, that Claude beats candidates on small coding tasks, is stated and not shown. No benchmark, no test set, no scoring, no example of a task where the model lost. Read as a manager reporting what he sees, it is believable. Read as a finding, it is unsupported.
What the post does not publish
How many people he has hired since dropping the test, or how they performed in the job. Whether Anthropic’s recruiting organization agreed, or whether other teams followed. And the part he admits: he has not settled on a replacement process, and is still scanning portfolios for prototypes while asking other managers for a better idea.
One more gap changes the size of the claim. A coding interview was a weak predictor, but a repeatable one: same prompt, same room, comparable scores. A weekend prototype is a better signal and a far less comparable one. It rewards the candidate with a free Saturday and enough confidence to build something nobody asked for.
There’s no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both.
On this page
What it changes for candidates and hiring managers
For a candidate, the preparation changes shape. Reading the company’s goals two levels up, picking one idea that would move them, and shipping something runnable beats another month of puzzle practice. The failure he warns about is scope: a prototype that carries a new language, a backend and mobile apps at once communicates nothing.
For a hiring manager, the post is a warning about what a prototype review measures. Taste and initiative are real. Free time and self-promotion come along with them, and they are not the same thing.
There is also a tension with Anthropic’s own numbers. The company’s employment report from October 16, 2026, which we covered here, put junior technical hires at 9% of technical hiring, down from 27% since early 2025. The essay says newcomers inside the company have an edge. The hiring data says fewer of them get in.
The Neural Minute take
The sentence people will quote is the slogan. The operative claim is quieter: the cost of trying fell, so a hierarchy built around expensive execution stopped making sense. That is testable in a way the post never tests. If a try costs a weekend instead of a quarter, the constraint inside a company is no longer engineering capacity, it is the decision about what deserves a try. No model prices that decision.
The one number in the piece is worth reading closely. Two engineers who were relatively new, Anthony Morris and Alice Zhao, built a graphical interface for Claude Code over one weekend. The first public version shipped roughly eight days later. That is a claim about how fast an idea can reach users. It is not a productivity measurement, and it says nothing about whether a planned build would have done better.
Our reservation is about the filter, not the argument. A hiring bar is a measurement, and this post replaces a bad measurement with an unmeasured one. The coding test was narrow and unpopular, but every candidate got the same task and a rejected one could ask why. A portfolio of weekend prototypes has no common task, no common rubric, no appeal. Rieseberg says as much in his last paragraph, where he asks other managers for a better method. The test was retired before the replacement was designed.
For anyone paying for an AI coding tool, the translation is direct. If the value is that a person can produce something demonstrable in two days, then seat counts and token spend are the wrong numbers. Watch how long it takes someone to get from an idea to something a colleague can click, and how often that prototype survives a real user.
Sources
- Felix Rieseberg, Hunger beats tenure: Advice for junior engineers, published October 27, 2026, read on October 28, 2026: https://felixrieseberg.com/hunger-beats-tenure-advice-for-junior-engineers/
- Role as stated in the post bio: engineering manager at Anthropic, led engineering for Claude.ai, Cowork and Claude Code Desktop.
- Figures quoted in the post: the Claude Code interface prototype built over one weekend by Anthony Morris and Alice Zhao, first public version about eight days later; the 1991 line from James March on naive newcomers; the 2011 line from Patrick McKenzie on business value.
- Anthropic employment report figures used for comparison: https://www.neuralminute.com/news/anthropic-junior-technical-hires-fall-to-9-percent/