An AI That Whispers Answers in Your Ear
Cluely is a startup selling real-time AI assistance that runs invisibly during video calls, job interviews, and sales meetings. The product displays a floating overlay – undetectable by screen-share software – that reads what is being said on the call and surfaces suggested responses, relevant facts, and talking points in real time. The pitch is simple: never be caught unprepared again. The backlash was equally simple: this is cheating, and the hiring industry is not ready for it.
The company launched with a marketing campaign built around controversy. Its founder, Roy Lee, was famously suspended from Columbia University for using AI to cheat on coursework, a detail the company leaned into rather than buried. Cluely’s promotional materials openly described the product as a tool to “cheat on everything.” That framing was deliberate – a provocation designed to generate press, and it worked. Within days of launch, the startup had accumulated a level of visibility that most seed-stage companies spend years trying to buy.
How the Overlay Actually Works
Cluely runs as a desktop application that sits beneath the visible layer of a screen-share session. When a user shares their screen on Zoom, Google Meet, or Microsoft Teams, the overlay does not appear in the shared view – it exists only on the user’s local display. The application captures audio from the call, transcribes it in real time using a large language model backend, and generates contextual responses that appear in a scrollable sidebar visible only to the Cluely user.
The technical execution is not entirely novel. Screen-invisible overlays have existed in gaming contexts for years, and AI transcription has been built into meeting tools at the enterprise level for some time. What Cluely did differently is combine these elements and point them explicitly at high-stakes evaluative situations: job interviews, technical screenings, and sales calls where performance directly determines income or career trajectory.
The product’s undetectability claim holds under most standard screenshare configurations, though some companies have begun testing full desktop capture requirements – forcing candidates to share their entire screen rather than a single application window. That countermeasure closes the gap partially, but does not solve the problem for voice-only interviews or phone screenings where no screen monitoring is possible at all.
The Hiring Industry’s Uncomfortable Position
Recruiters and hiring managers are left without a clean answer here. Traditional integrity tools – plagiarism detectors, timed take-home tests, live coding environments – were built for a world where AI assistance was slow, clunky, or obviously visible. Cluely and tools like it operate faster than a candidate could realistically research an answer manually, and the output is indistinguishable from genuine recall.
Some enterprise hiring platforms are responding by shifting toward evaluation formats that resist this kind of assistance. Behavioral interviews structured around unpredictable follow-up questions, whiteboard-style problem solving conducted over video with camera-on requirements, and multi-round human conversations are harder to game with a sidebar tool. But smaller companies and early-stage startups that run lean recruiting processes – often a single video call and a take-home – have almost no structural defense against it.
Who Is Actually Using This, and Why
The user base Cluely is targeting is not a fringe category. It is candidates who are already exhausted by an application process that has itself leaned heavily into automated screening. Companies routinely filter resumes using AI tools, send automated rejection emails, and deploy asynchronous video interview platforms that record candidates answering pre-set questions with no human present. When the hiring process becomes that depersonalized on the employer side, the moral weight of using an AI assist on the candidate side gets harder to argue cleanly.
That tension is real, and Cluely’s founders appear to be deliberately exploiting it. The framing of the product as leveling an uneven playing field has found traction among job seekers who feel the current process is already stacked against them – particularly those who are not native English speakers, those with interview anxiety disorders, or those returning to the workforce after long absences. For this group, a real-time prompt suggesting how to articulate an answer they actually know is meaningfully different from an AI generating an answer they do not.
The ethics, though, do not stay that clean once the tool scales. A candidate with interview anxiety using Cluely to surface a phrase they were struggling to recall is a different situation than a candidate using it to fabricate domain knowledge they simply do not have. The product makes no distinction between these uses. It responds to whatever the interview question is and generates the most useful-sounding answer available, regardless of whether the user actually possesses the underlying competence being evaluated.
That gap between performance and competence is where the downstream consequences land – not in the interview room, but on the job. A candidate who passes a technical screening with Cluely’s help and lands a role that requires that technical knowledge does not suddenly acquire the knowledge. The deception moves from the hiring process into the workplace, and the employer absorbs the cost. How long before companies start building post-hire assessment periods specifically designed to audit whether a new employee’s skills match their interview performance?
