Aakash GuptaCollege Dropout Raised $20M Building AI Tools | Cluely, Roy Lee
CHAPTERS
How Cluely decides what to build: ship broad, let usage data pick winners
Roy explains that prioritization comes from sheer volume of real-world usage: millions of daily requests and hundreds of customer emails make pain points and sticky use cases obvious. Instead of heavy roadmaps, the team launches a general tool, watches how people use it, then iterates rapidly toward what works.
Roy’s origin story: provocative by default, rewarded and punished
Roy describes being polarizing from early childhood—saying thoughts publicly and brazenly—which created both strong supporters and strong detractors. He frames this trait as the through-line that later shaped his unconventional path and willingness to take bold bets.
Harvard rescinded and the Columbia chapter: school as optional
Roy recounts getting accepted early to Harvard, then losing it after being mass reported and suspended. He connects that experience to rethinking the value of school and channeling his ambition into building a company instead.
10 weeks old, ~$6M ARR: why Cluely feels everywhere already
Roy shares that Cluely is only about 10 weeks from the first line of code, yet is nearing $6M ARR and constantly on the timeline. He attributes the mismatch between company age and perceived maturity to an unusually strong marketing/distribution engine.
Becoming the main character on X: applying short-form virality rules to tech
Roy argues that X/LinkedIn creators over-index on sounding smart, while algorithms reward digestible, controversial content that forces a reaction. He claims he simply applied obvious short-form playbook tactics to Tech Twitter and scaled quickly.
Controversial content that sells: stop over-optimizing funnels, chase attention
Roy downplays precise conversion tracking, claiming attention and brand visibility are the scarce asset—especially early. He emphasizes that viral formats change fast, so teams must iterate constantly rather than rely on stable, repeatable funnels.
A repeatable distribution machine: UGC at scale + daily stunt brainstorming
Roy details Cluely’s content operations: viral-sense staff, daily ideation, and a pipeline that turns cultural observations into stunts and formats. He claims they generate dozens of high-upside ideas per day and execute quickly across platforms.
Execution infrastructure: 60+ creators, influencer outreach, building a film studio
Roy explains how Cluely bridges the gap from idea to output by building internal systems: a large retainer-based creator bench, influencer outreach processes, and an emerging in-house studio for “movie-quality” weekly launches. When they can’t do a format internally, they outsource and later consider bringing it in-house.
Product deep dive: translucent overlay UX (Interview Coder → Cluely)
Roy positions the translucent overlay as the “true GUI for AI,” eliminating split-screen chatbot friction by integrating assistance into any workflow. He says Interview Coder served as the prototype, with 20–30 iterations before landing on the final seamless design, led by founders rather than dedicated designers.
Lean team, user-led roadmap: four engineers and fast alignment
Roy says coordination is simple because the team is tiny—only four engineers—so decisions happen in the room without heavy process. The company stays oriented around user behavior and inbound feedback, not sprint rituals, and leverages its distribution to attract strong talent.
Core technical approach: audio capture + screenshot-at-query + latency workarounds
Roy outlines the technical constraints of real-time “screen + audio” assistance. Instead of continuous video context, Cluely captures screenshots at query time, compresses images to reduce tokens, and uses a custom audio engine for system + mic capture, while exploring server optimizations and potentially self-hosted models to reduce latency variance.
Model strategy and the application-layer bet: SaaS margins, not AGI burn
Roy says Cluely primarily uses OpenAI (GPT-4.1) today and expects a multi-model future like other AI apps. He argues application-layer companies can capture enormous economic value without the massive capital burn required to compete at the frontier-model/AGI layer.
Enterprise traction and the ‘Cluely for sales’ angle: coaching via post-call summaries
Roy highlights enterprise momentum and features built for large customers, especially post-call summaries that identify missed moments where Cluely could have helped. He frames this as a training/coaching tool for sales reps while maintaining a broad, default-AI-for-everyone ambition.
The long-term vision: new AI GUI everywhere, CRM displacement, and brain chips
Roy predicts chatbot UX will be obsolete as models improve and become context-aware of screens and audio. He paints two end states: for enterprise, a real-time AI-native system that pulls from and writes back to CRMs; for consumers, ultimately brain-computer interfaces as the best AI form factor.
Working at Cluely: frat-house culture, extreme commitment, and extreme comp
Roy describes an intense live-work environment in a shared SF mansion where the company becomes employees’ social world. He argues high dedication requires high pay and equity, and claims the culture self-selects for people who want that lifestyle—while acknowledging possible future downsides.
Ethics, legality, and ‘are we getting dumber?’: Roy’s defense of AI cheating tools
Responding to concerns from educators and public critics, Roy argues that technology always obsoletes certain knowledge and that this is progress, not decline. He dismisses jail rumors as misreadings of his online persona and reiterates that people underestimate how new and iterating the product is.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome