From expelled student to $5M ARR: Roy Lee and Cluely

Executive overview

Technical interview prep is a broken ritual — developers memorise hundreds of algorithm puzzles that have nothing to do with actual work. Roy Lee built a tool to cheat on those interviews, got expelled from Columbia for it, and used the viral moment to launch Cluely: an AI screen overlay that assists with anything in real time.

The broader bet is that the chat-prompt interface will be replaced by ambient AI that sees and hears your context. Cluely is a proof of concept for that future — and the "cheat on everything" framing is deliberate provocation to make people question where the line is.

When AI can do something for everyone, it stops being cheating and becomes the new baseline.

The interview coder origin

• LeetCode is 300–600 memorised riddles used to gate $200k jobs — it tests memorisation, not engineering skill.
• Roy built a tool that invisibly overlays AI assistance during technical interviews.
• He recorded himself using it on an Amazon interview, got the offer, and posted the video publicly.
• Virality took a month to arrive; during that time co-founders wanted to quit, but he held on.
• Going viral provided cover from Columbia's punishment and opened the path to fundraising.
• The product was designed to die: once companies change their interview format, it's obsolete — so he had to move fast.

Building Cluely

• Cluely is a translucent desktop overlay that sees your screen and hears your audio — a novel UX that had never been attempted.
• The core insight: multimodal AI that knows your context without prompting is where the interface is heading.
• "Cheat on everything" is intentionally ambiguous — the friction people feel is a gut reaction to asymmetric advantage, not genuine harm.
• The main technical challenges are latency and accuracy, not the underlying concept.
• Plans to self-host models to cut latency from OpenAI load balancing overhead.
• Long-term moat: hyper-personalised fine-tuned models built from each user's accumulated interaction data.

Going viral on purpose

• Twitter rewards controversy; every planned viral tweet needs a strong negative reaction from roughly half the audience.
• Interview Coder landed well because it read as a scrappy underdog fighting big tech — framing matters.
• As the account grew, the same content received less support — virality resets with audience growth.
• Roy separates his online persona (deliberately provocative) from his real identity — the people who matter are a small, trusted circle.
• All press is good press in practice: negative comments drive downloads and awareness for the core product.

Risk-taking as a repeatable process

• Risk compounds: start with a small action (build but tell nobody), then iterate to bigger exposure.
• Limiting beliefs about post-graduation paths (engineer, lawyer, doctor) are the first thing to discard.
• The downside of a risk is almost always smaller than imagined; the upside larger.
• AI lowers the bar to expertise — two months of study can make a 19-year-old a genuine pioneer in the field.
• Taking bigger risks forces resourcefulness; the capability was always there, the constraint was permission.