By Lee Flanagan
✨ AI Summary:
- In-person interviews alone do not prevent AI-assisted cheating; the real problem is static test design with one retrievable correct answer.
- Shift from closed-answer coding tests to project-based exercises that measure how candidates work with AI tools and adapt to real-time challenges.
- Trained interviewers who probe reasoning, challenge assumptions, and follow up on inconsistencies create reliable signals regardless of interview format (video or in-person).
- Rebuild what interviewers ask and how they evaluate responses rather than adding logistics costs like mandatory office visits, which buy a room but not better hiring outcomes.
“Are you interviewing Claude or the person?” That’s the question Nick Bloom, the Stanford economist known for his remote-work research, put to the Financial Times’ Economics Show podcast, in an episode released on Friday and reported by Business Insider. His answer, in effect: tech firms cannot tell anymore, so they are dragging candidates back into conference rooms to find out. That response treats the wrong variable. The problem was never the webcam. It is the question on the other side of it.
The Coding Test Was Already Broken
Look at what Bloom actually describes. “A lot of tech firms used to hire folks based on coding tests. They can’t do that anymore,” he told the FT, because candidates can run the prompt through an AI tool and read the answer straight off the screen. That requires a question with one retrievable correct answer, delivered in a format that rewards recitation over reasoning. That is a test-design failure. Move the same candidate into a conference room and hand them the same static coding problem, and a well-prepared answer memorised in advance works just as well as one read off a hidden screen. The room changed. The vulnerability did not.
Greenhouse’s Own Numbers, Read Correctly
There is real movement behind Bloom’s claim, worth naming at scale. Greenhouse’s 2025 AI in Hiring Report, based on more than 4,100 job seekers, recruiters and hiring managers across the US, UK, Ireland and Germany, found that 39% of US hiring managers were conducting more in-person interviews “to verify” candidates. Greenhouse’s 2026 report, with a smaller sample covering only the UK, Ireland and Germany, shows the same pattern continuing.
Read that as what it is: a vendor’s own survey of practitioner behaviour, not proof that the behaviour works. A hiring manager choosing to “verify” in person tells you plenty about anxiety and very little about outcome. Our read is that the 39% figure captures a defensive reflex spreading through the market faster than any evidence that it actually stops AI-assisted cheating. Adding a flight and a badge check to your process is friction. Friction is not a fix.
Cisco Chose Method Over Geography
Compare that reflex to what Cisco is doing. Business Insider reported that Cisco is moving from conventional coding challenges toward project-based exercises that observe how candidates operate inside AI-enabled workflows. Scott McGuckin, Cisco’s VP of global talent acquisition, put it plainly: “The human element of oversight and expertise is more crucial than ever.” Google has piloted letting software engineering candidates use an approved AI assistant during interviews. Vibe-coding startups including Cognition, Base44 and Replit have embraced AI use in technical hiring.
None of these employers solved the problem by choosing a location. They solved it by changing what the exercise measures. A project-based task that watches how someone works with AI tools, adapts to a wrinkle mid-task, and explains a tradeoff on the fly cannot be defeated by reading an answer off a screen. There is no single answer to read. Cisco’s shift measures how someone works, not what they can recall.
A Scripted Question Is Gameable Anywhere
Here is the challenge worth putting to any TA leader currently weighing the cost of mandatory in-person final rounds: what exactly does the room stop your interviewer from doing wrong? If your interview still asks a closed question with a memorised or AI-generated right answer, in person just slows the cheating down. It does not eliminate it. A candidate who has rehearsed the answer to “tell me about a time you resolved a conflict” will deliver that answer just as smoothly across a table as across a screen. What stops them is a trained interviewer who follows up, probes the reasoning, asks why the candidate chose that approach over another, and pushes on the parts of the story that do not add up. That skill travels. It works on Zoom. It works in the room. It is the one thing AI cannot pre-load for a candidate, because it depends on what happens after the first answer, not before it.
What we see across TA teams is a rush to solve an evaluation problem with a logistics decision, an expensive way to land on the same weak signal you started with. Money spent flying candidates back to the office buys you a room. Money spent rebuilding what your interviewers ask and how closely they probe an answer buys you a signal that holds up anywhere a candidate sits. Bloom has identified a real failure in how tech companies evaluate candidates. Sending people back to the office does not fix it. Rebuilding the question does, whatever room it gets asked in.
Original reporting: Business Insider.
Frequently asked questions
What did Nick Bloom mean when he asked whether firms are interviewing Claude or the person?
Bloom, a Stanford economist, was describing how candidates can run coding test prompts through an AI tool and read the answer straight off the screen during a video interview, which is why some tech firms cannot tell if they are evaluating the candidate or the chatbot.
Does Greenhouse’s 39% figure prove in-person interviews stop AI-assisted cheating?
No. It shows that 39% of US hiring managers were adding in-person interviews to verify candidates, which is a record of practitioner behaviour, not evidence that the behaviour actually prevents AI-assisted cheating.
How does Cisco’s project-based interview differ from a standard coding test?
Instead of a closed problem with one retrievable answer, Cisco watches how candidates operate inside AI-enabled workflows, with VP of global talent acquisition Scott McGuckin telling Business Insider that human oversight and expertise ‘is more crucial than ever.’
What stops a candidate from reciting a rehearsed answer once they are sitting across the table?
A trained interviewer who follows up on the answer, asks why the candidate chose that approach over another, and pushes on inconsistencies produces a reliable signal, whether the interview happens on a screen or in a room.