Briefing
Episode 34 · Primer and discussion · About 15 minutes
Download the MP3 · Linda and Leonard, AI voices, read a script written from the sources below.
Download the MP3 · Linda and Leonard, AI voices, read a script written from the sources below.
Checked against Rieseberg's post and Anthropic's own guidance on 2 November 2026. Claims about what Claude can do and about the prototype timeline are his account. Check the original post before quoting it.
Before we record, every claim the hosts make is checked against the sources (this table was last updated 2 November 2026). Confirmed means a primary source or two independent reports; Reported means one party says so; Partly right, Disputed, Unverified and Not yet known mean what they say, and the hosts say so on air. This is the table as we left it.
Sign up for our podcast at agile-insights.com.au.
Linda Here is the idea for today. When A I can do the task a test was built to check, the test stops telling you who is good. So an organisation has to rethink how it judges people, and who gets to back an idea, not just make the old work faster. By the end you will know why one Anthropic manager dropped the coding interview, and what that says about how your organisation measures people.
Leonard Let me start with what happened, because the headlines ran ahead of it. On the twenty seventh of October, Felix Rieseberg, an engineering manager at Anthropic, published a blog post called Hunger beats tenure. He has led engineering for Claude dot A I, Cowork and the Claude Code desktop app. He wrote that this month, as a hiring manager, he stopped asking candidates to do traditional coding interviews. One newsletter headlined it as Anthropic ending coding interviews. That is not what the post says. It is his own hiring. Anthropic's published guidance for candidates still says live interviews are done without A I, unless the company says otherwise.
Linda Good, so let me build the concept up in four steps. Step one is what a test is for. A coding interview is a stand in. Nobody hires you to solve a puzzle in forty five minutes. They use the puzzle because your score is meant to line up with how well you will work later. Rieseberg called it a weak but useful predictor on his teams. The word that matters there is predictor. A test earns its place only while its scores separate the people who will do well from the people who will not.
Leonard Before you go on, let me ask the obvious newcomer question. If the test worked for years, what broke it? Did the candidates get worse at coding, or did something else change?
Linda Something else changed, and that is step two. The tool got better. Rieseberg says there is no more point in a little React challenge than in quizzing people about assembly, because Claude is now better at both. That is his judgement, and he gives no measurement for it. But follow the logic. If a machine can do the task well, anyone with the machine scores well. When everyone scores well, the score stops separating anyone. He says his most impactful builders succeed regardless of how they do on these tests. The test has lost its signal.
Leonard So the machine sits the exam.
Linda And passes it.
Leonard That sounds like it reaches well beyond programmers. Plenty of what organisations use to judge people is a stand in like that. A certification, a writing sample, years in the role before a promotion, a reviewer's sign off on a report. If A I now supplies the skill underneath, do those go the same way?
Linda That is our reading, not his finding, and it is worth testing case by case. Step three is where the bottleneck moves. Rieseberg argues that companies are built around a few people who green light or block ideas, because doing the work used to be expensive. You had to earn status before the organisation would take a bet on you. His line is that execution is not expensive any more. His example is the graphical version of Claude Code. By his account, two relatively new engineers, Anthony Morris and Alice Zhao, built a working prototype over one weekend. The first public version shipped roughly eight days later...
Leonard Step four is the one in his title. He says the word junior is wrong. His advice is really for engineers without social capital, status or power. He quotes the organisational theorist James March, who wrote in 1991 that knowledge depends on maintaining an influx of the naive and ignorant. Newcomers notice the odd things that old timers have stopped seeing. The old system often wasted that, because a newcomer's idea needed a sponsor. Rieseberg says he now sees comparatively new engineers fly past better coders who have been around much longer. Hunger beats tenure. That is what he sees on his own teams, not the result of a study.
Linda Now the common misunderstanding. The lesson is not that skill no longer matters, and it is not that Anthropic has given up on testing people. Rieseberg says plainly that he has not found a replacement yet. For now he scans résumés and portfolios for small prototypes that show taste, empathy and principled thinking, and he asks other managers for better ideas. Dropping a test with nothing tested behind it is a gap, not a reform. And notice that one manager's practice and the company's written guidance now say different things.
Leonard So the measure has broken, the written policy and the real practice have drifted apart, and the replacement is not built. For someone responsible for governing A I in an organisation, what does all that add up to?
Linda It adds up to this. A I does not only speed up existing work. It quietly breaks the measures you use to judge work and people. So governance has three jobs here. Find the assessments and controls that assume a skill A I now supplies, and ask whether each one still predicts anything. Give good ideas from people without status a route to someone who can say yes. And make that route safe, with data, security and release rules, so that speed does not become a way around them.
Leonard And one more check that costs almost nothing. Compare what your written policies say with what your managers are actually doing. The gap we saw at Anthropic, between published guidance and one manager's hiring, probably has a version in yours. Better to find it on purpose than in an audit. In Part 2 we argue about what should replace the test, and whether hunger really does beat tenure. Thanks for listening.
Linda Here is the idea for this discussion. When A I can do the task a test was built to check, the test stops telling you who is good. So an organisation has to rethink how it judges people, and who gets to back an idea, not just make the old work faster. By the end you will have three answers to what should replace a broken test, where the evidence settles, and what to do on Monday.
Leonard A quick recap for anyone who skipped Part 1. On the twenty seventh of October, Felix Rieseberg, an engineering manager at Anthropic, wrote that he had stopped giving candidates coding interviews. His reason is that Claude is now better than people at the small tasks those interviews use, so the test no longer separates anyone. That is his judgement, with no measurement attached. It is his own hiring, not company policy. Anthropic's published candidate guidance still says live interviews are done without A I unless the company says otherwise. And he says he has not yet found a replacement.
Linda So here is the question. Organisations judge people by tests of skill all the time. Interviews, certifications, promotion criteria, a reviewer's sign off. When A I can pass those tests, what should take their place? I want to give three answers real room. Drop the test and judge people by what they have built. Keep a test but let A I into the room. Or stop worrying about the gate at the front door and fix how ideas get to a decision once people are inside...
Leonard I will start with the first answer, and with my worry about it. Rieseberg's stand in, for now, is scanning résumés and portfolios for small prototypes that show taste, empathy and principled thinking. I like the target. But think about who has a portfolio of weekend prototypes. People with spare weekends, good equipment, and someone telling them a portfolio matters. That can become a status filter by another name, which is the very thing his post argues against. His advice is aimed at people without social capital, status or power. A portfolio screen may quietly favour the people who already have them.
Linda That is a fair warning, and it points to the second answer. Do not drop the test. Change it so it measures the work as it is now done. Google is the example people cite. According to reporting by Business Insider, Google has been piloting coding interviews since May in which candidates may use its Gemini assistant. The format was reportedly expanded in October. Interviewers look at how the candidate directs the tool and checks what comes back. That keeps a structured test, the same for everyone, but it moves the measure from typing the answer to judging it. Judgement is the skill Rieseberg says is now scarce.
Leonard I will grant that it is fairer than a portfolio. My concern is the evidence. A test earns its place by predicting later performance, and nobody has published whether the A I assisted format predicts anything yet. We would be swapping a test we know has lost its signal for one we have not measured. That is not a reason to wait. It is a reason to treat the new format as a pilot, not as the answer. Check its results against how people actually perform after a year.
Linda Agreed, and that is the governance habit in one line. Pilot it and measure it. Now the third answer, which is where I think the post is strongest. Rieseberg's real complaint is not about interviews. It is that companies are built around a few people who green light or block ideas, because the work used to be expensive. So a newcomer's idea waited for a sponsor. His point is that execution is cheap now. His example is the graphical Claude Code. By his account, two relatively new engineers built a working prototype over a weekend, and a first public version shipped roughly eight days later. The lever there is decision rights, not hiring...
Leonard This is where I part company with you a little. A weekend prototype skips the requirements document, the buy in and the queue for design time. It can also skip the security review, the data rules and the question of who supports it. Rieseberg is describing people moving fast inside a company that builds these tools for a living. In most organisations, that same story is shadow I T. Someone connects a prototype to real customer data on a Saturday, and by Tuesday it is something people rely on that nobody approved. Speed with no route through the controls is a bypass.
Linda I accept that, and it sharpens my answer rather than ending it. The aim is not to remove the gate. It is to make the gate quick and safe. Give people a sanctioned place to build, with test data, the approved tools and clear rules about what can touch production. Then give them a regular slot to show the prototype to someone who can actually say yes. That way the newcomer's idea reaches a decision in days, and it reaches it through the controls rather than around them. Fresh eyes without a route to a decision are wasted, and so is a route that skips the controls.
Leonard Then we land in the same place. I would add one reason the third answer matters to governance leads in particular. Rieseberg quotes James March, who wrote about this in 1991. March argued that organisations which refine what they know faster than they explore new things do well in the short run. In the long run they harm themselves. March wrote that knowledge depends on an influx of the naive and ignorant. A I makes exploring cheap. The organisations that benefit will be the ones whose decision process can absorb more experiments, not the ones that only make the existing backlog faster.
Linda Well put.
Leonard I have my moments.
Linda So where does the evidence settle? A few things are confirmed from the primary sources. The post is one manager's practice, not a company policy. Anthropic's written guidance, last updated in July twenty twenty five, still keeps A I out of live interviews unless the company says otherwise. And Rieseberg says himself that he has no standard replacement. Several things are not known. Whether Claude really beats candidates at these tasks is his judgement, not a measurement. Whether newer engineers outperform at scale rests on what he has seen on his own teams. And the weekend and eight days story is his account; we found no dated release note to set beside it.
Leonard And two predictions are still only predictions. He says he does not think the coding interview is coming back. And our own reading is that this reaches past engineering into promotions, certifications and sign offs. That is an inference for each organisation to test, not a finding. Nobody has published data on whether the new formats predict performance better than the old ones. So the honest position is this. The old test is weaker and the new tests are unproven. The people who measure carefully over the next year will learn the most.
Linda So here is what an organisation can do. First, list every place you judge people or work by a skill that A I now supplies. That runs from hiring tests to promotion criteria to review sign offs. Ask of each one whether it still predicts anything. Second, where it does not, run a replacement as a pilot and check it against later performance. Do not drop the old test with nothing behind it. Third, give ideas from people without status a quick, regular route to someone who can say yes, inside a safe place to build.
Leonard And fourth, compare what your written policies say with what your managers are actually doing. The gap between published guidance and one manager's hiring is the clearest lesson in this story. You will probably find a version of it in your own organisation. Find it on purpose, decide which side should move, and write the answer down.
Linda If you'd like to know more, sign up for our podcast at agile-insights.com.au.
Leonard Thanks for listening.
All of the briefings (38) · Follow Briefing in Apple Podcasts or any podcast app
Everything from Agile Insights: briefings, Making You Think and the reference series