Inside the AI Gold Rush: Cluely Founder Roy Lee on Making Money, Losing Jobs, and What Comes Next
Roy Lee, the 21-year-old founder of the AI company Cluely, has built a reputation as one of tech’s more polarizing figures — first for a stunt that got him removed from Columbia University, and more recently for running a company that operates with a strikingly small internal team while generating outsized revenue. In a recent podcast appearance, Lee laid out his view of where the money is being made in AI right now, why he thinks entry-level knowledge work is disappearing, and how his own company is built around a model most startups haven’t adopted yet.
The New AI “Gold Rushes”
Lee’s central argument is that most of the money to be made in AI right now isn’t in building the underlying models — it’s in exploiting the gap between what AI can already do and what the average business or consumer realizes it can do. He estimates the overwhelming majority of companies today have some obvious application of AI they simply haven’t implemented yet, whether that’s advertising, web development, or customer-facing products.
He points to AI-generated video as the clearest low-effort entry point. His pitch: use an AI video generation tool to produce a sample ad for a company, email it to them for free, and offer a paid retainer only if the ad performs. If it doesn’t work, the company owes nothing and never hears from you again. Lee argues this model is spreading fast — including, he claims, agencies that quietly outsource the actual video creation to teenagers and resell the output as professional marketing work.
He breaks the opportunity into rough tiers. At the top, for people with strong technical backgrounds, is work that directly serves AI labs — supplying training data, building infrastructure, or improving chip and GPU efficiency — a category he says is flush with venture capital. Below that is app development: building and marketing a consumer app, something he argues has never been cheaper or faster to do, aided by AI-generated marketing content. At the more accessible end is exactly the ad-creation approach described above.
His practical playbook for scaling that lowest tier: identify well-funded software companies, produce a genuinely compelling ad for each one using an AI video tool, and cold-email them daily with a no-risk offer — free ad, paid retainer only if it performs. Repeated across enough companies, he argues, this can realistically generate significant recurring revenue within months. He also suggested a simpler shortcut: find a company’s best-performing existing ad on a public ad library, recreate it with AI tools, and pitch it to their competitors.
Why Lee Thinks Younger People Have an Edge
Lee’s argument for why younger, terminally-online people are unusually well positioned in this landscape is somewhat unconventional: years spent consuming huge volumes of short-form video content has trained an instinct for what feels “boring” versus what will hold attention — a skill he says is unevenly distributed across age groups and hard to fake. He argues that instinct, more than any formal skill, is what’s needed to iterate an AI video generation tool toward content that actually performs as an ad.
How Cluely Operates Differently
Cluely, Lee’s company, currently maintains seven software products under one corporate entity — including its flagship desktop app — with only two full-time engineers. Lee attributes this to the leap in capability of AI coding tools, which he says has narrowed the gap between novice and senior engineers enough that small teams can now maintain what would previously have required a much larger engineering organization.
The other distinguishing piece of Cluely’s model is its approach to marketing: rather than a conventional in-house creative team, the company works with what Lee says is over a thousand performance-based content creators, paid only when their videos succeed as ads — with room, he claims, to scale that further. He argues most companies haven’t caught onto this approach yet, seeing an open opportunity in simply outproducing competitors on organic video content. He was skeptical, however, that this model could be turned into a scalable B2B consulting service, arguing that marketing is fundamentally a zero-sum competition for a fixed amount of attention, unlike efficiency gains in something like payroll processing that benefit everyone simultaneously.
The Future of Entry-Level Work
Lee was blunt about his expectations for the job market facing new college graduates, estimating — without citing hard data — that a meaningful share of this year’s graduating class won’t find employment in their field, and arguing the situation will worsen considerably over the next five years as AI tools increasingly handle work previously done by junior employees. Despite that outlook, he said he’d still advise teenagers to stay in high school rather than drop out to pursue business full-time, arguing that the social and developmental value of that environment outweighs the head start an early exit might provide — citing his own experience with a year of post-graduation isolation as a formative, difficult period.
AI, Loneliness, and Companionship Apps
Asked whether AI is contributing to male loneliness, Lee pushed back, attributing most of the decline in in-person socializing to the pandemic and the broader rise of social media rather than AI specifically. He pointed to shifts in how people meet partners — increasingly through dating apps rather than in-person environments — as a downstream effect of the same forces.
On AI companion apps specifically, Lee offered a pragmatic, if debatable, take: for someone who is otherwise isolated and unlikely to change course on their own, he argued an AI companion is preferable to nothing at all, while acknowledging present-day versions of these products are limited. His hope, he said, is that future versions could eventually help nudge isolated users toward real-world social connection rather than substituting for it entirely.
On AI Safety and Public Perception
Lee also addressed the broader public conversation around AI risk, offering the view that heightened fear of AI can financially benefit certain companies by discouraging engineering talent from working on competing frontier models — while acknowledging he has no way to confirm whether any specific instance of public alarm was deliberately engineered for that purpose. He discussed the temporary restriction and later restoration of Anthropic’s Fable models, characterizing the underlying concern as being about model capability generally, without going into technical specifics.
Asked about reports of tech billionaires building private bunkers, Lee framed it as a response to a scenario in which sufficiently advanced AI capabilities end up in the hands of an actor willing to cause catastrophic harm, though he said he personally hasn’t taken similar precautions.
Life Philosophy and Personal Risk
The conversation closed on more personal territory, with Lee describing an intensely ambitious, risk-tolerant approach to his own life — including a willingness to accept significant personal cost, financial and physical, in pursuit of outsized outcomes. He was candid about experimenting with various performance-enhancing substances and about the toll he expects some of those choices may eventually take, while framing that risk tolerance as central to how he’s approached both his career and his personal life.
The Bigger Picture
Setting aside the more provocative moments of the conversation, Lee’s core thesis is a fairly direct one: AI has dramatically lowered the barrier to building products, running lean teams, and producing marketing content, and most companies and individuals haven’t caught up to that shift yet. Whether or not one agrees with his specific tactics or his broader worldview, the underlying observation — that there’s currently a wide gap between what AI tools can do and what most businesses have actually implemented — is one echoed by a growing number of people building at the intersection of AI and small-team entrepreneurship.
