He built an AI app to beat coding interviews. Then Columbia suspended him

A software application called Interview Coder promises to help software developers succeed at technical job interviews—by surreptitiously feeding them answers to programming questions via AI.

Interview Coder’s 21-year-old cofounder and CEO Roy Lee says he and Neel Shanmugam, the company’s cofounder and COO, created the tool partly as a protest against longstanding industry practices that require job candidates to solve programming puzzles during interviews. Lee, who until recently was a sophomore at Columbia University, says he spent hundreds of hours practicing such problems—time that could have been spent on actual coding projects.

“This kind of killed a lot of my love for programming, just because I was forced to write code that just wasn’t fun,” he says. “I was forced to solve riddles instead of actually working on building real world projects, and I just grew to really dislike the system.”

But protest or not, Lee says the project has proven lucrative, recently surpassing $3 million in annual recurring subscription revenue—presumably from customers more interested in cheating their way into a job than making a statement. The program, available for Windows and Mac, allows users to secretly take screenshots of programming puzzles presented during interviews, feeding the questions to AI for analysis and coded solutions.

The software is designed to evade detection by anti-cheating measures in interview platforms. Controlled by keyboard shortcuts, it avoids the giveaway of mouse movements. Interview Coder can even be placed transparently atop the interview window, so users don’t appear to shift their gaze while consulting AI-generated solutions and talking points. Lee recommends that users practice with the tool before deploying it in a real interview.

Lee says he personally tested the software in real internship interviews and has posted videos online that appear to show him using Interview Coder during a challenge for Amazon. “We posted videos of me using it on Amazon, primarily, which is like the big boss interview that we took down,” he says.

According to Lee, that led to takedown attempts by Amazon and disciplinary action from Columbia. He says the university initially placed him on probation over concerns that the tool could be used to cheat on class exams, then suspended him for a year for recording a disciplinary hearing and sharing related documents without permission. Lee says he’s unsure if he’ll return to school.

He has published marked-up versions of Columbia documents related to the matter. The university declined to comment, citing federal privacy law, and Fast Company was unable to independently verify the authenticity of the documents.

Amazon also declined to comment on Lee or his application but said, through spokesperson Margaret Callahan, that candidates are generally asked to acknowledge they won’t use unauthorized aids like generative AI during interviews and assessments.

For his part, Lee believes big tech interview processes should better reflect actual working conditions. “It doesn’t make sense to test someone on riddles and essentially give them an IQ test when they’re not going to be doing that at all,” he says.

In his view, job candidates should be allowed to use any tools they’d have access to on the job—including AI—during interviews.

Having an AI copilot during an interview isn’t a completely far-fetched concept. Some technical interview platforms, like CoderPad and CodeSignal, already allow clients to enable AI assistants for candidates. But, says CodeSignal CEO Tigran Sloyan, that typically involves redesigning interview questions to suit AI usage and even revising job descriptions to reflect that AI proficiency is part of the job. He likens the transition to the introduction of digital calculators in schools, which eventually led to rethinking how math was taught and tested.

It’s also critical, Sloyan adds, that companies provide the AI tools themselves rather than letting candidates bring their own.

“Right now there is a gigantic menu of all sorts of different AI tools, some of which are very expensive to use, and there is a significant difference in what they can and cannot do,” he says. “So especially in the hiring process, when you want to give everybody a consistent and fair shot, the AI has to be embedded in the interview and assessment platform itself.”

Sloyan also suggests Interview Coder may not be as stealthy as its creators claim. Beyond the technical countermeasures platforms like CodeSignal use to detect cheating, candidates might also give themselves away by awkwardly reading answers from a hidden window—behavior that differs from natural brainstorming.

“I would highly encourage candidates who are going through an assessment process with CodeSignal to think twice before using something like Interview Coder, because we flag it countless times, and the claim that it’s completely undetectable is not true,” he says.

Lee acknowledges that, at least for now, his software might help people cheat into jobs they wouldn’t otherwise land—or get caught trying. “Sure, there will be some bad eggs that get caught or that just slip in,” he says.

But Lee—who’s exploring other applications for screen-aware AI assistance—argues that Interview Coder is ultimately meant to make itself obsolete by pushing companies to modernize how they evaluate candidates.

“The product is meant to kill itself,” he says. “And the day it kills itself is the day that every single company will hire significantly better engineers, and every engineer will be better because they’ll spend more time engineering instead of [solving programming puzzles.] It’ll just be a huge net positive for the developer community.”

https://www.fastcompany.com/91317966/interview-coder-columbia-suspension?partner=rss&utm_source=rss&utm_medium=feed&utm_campaign=rss+fastcompany&utm_content=rss