Cluely founder Roy Lee said on the Jack Neel Podcast that three small-scale "gold rushes" are unfolding in AI and are still not being fully used: supplying infrastructure to large AI labs, quickly building and marketing AI apps, and offering AI video and advertising services.
Lee said the most attractive mass-market opening is not training models from scratch or launching an AI company that needs a large engineering staff. In his view, the real opportunity is to use the latest generative AI tools to make money from people and companies that still do not realize how usable those tools have become.
"99.9% of companies in the world have some area where they could use AI, but they still haven’t done it," Lee said.
First wave: serving AI labs through infrastructure
For founders with strong technical ability and deep engineering backgrounds, Lee said the best target remains the frontier AI labs, including OpenAI, Anthropic, and Google DeepMind.
He pointed to data services, model infrastructure, GPU and compute systems, and any technology that can improve chip efficiency or model execution. AI labs are spending billions of dollars to build next-generation models, he said, and solving one important bottleneck in that stack can create a large B2B business.
Lee described this layer as the closest thing to "selling shovels" in the current AI rush. He also said it carries the highest technical barrier, which means it is not open to everyone.
Second wave: AI apps can now be built far faster
For people who cannot compete in AI infrastructure, Lee said the next opening is to build AI applications directly.
He argued that coding agents and code-generation models have compressed software development timelines. Products that may once have required 20 engineers and three to six months to complete can now, in some categories, be prototyped by one engineer in a very short period of time.
Using Cluely as an example, Lee said the company maintains seven software products with only two full-time engineers. He also claimed that several of those products already produce more than $1 million in annual revenue.
The article noted that those revenue figures and staffing details came from Lee’s comments on the podcast and that no independent financial data is currently available to verify them.
The broader point, he said, is that AI changes product strategy. Instead of putting all company resources behind a single product, startups may be able to run many low-cost experiments.
In the older model, launching a new product meant hiring a team, writing code for months, designing the UI, testing, and deploying. Now, Lee said, the process can look more like building seven products first and then seeing which one finds market fit.
That shifts the structure from "one company, one core software product" to "one company, one very small team, multiple AI software experiments." Lee said that may become a common operating model for AI-native companies.
Third wave: AI ads as the lowest-barrier trade
Lee said the lowest-barrier and most underestimated AI business right now is AI video.
His reasoning is tied to distribution. TikTok, Instagram Reels, and YouTube Shorts have become central arenas for attention, while AI video generation is rapidly lowering the cost of producing ads and UGC, or user-generated content.
Because of that, he said, a person does not even need to start an AI company first. The simpler move is to make ads for existing companies.
Lee laid out a direct playbook: identify 100 software companies that have raised more than $10 million, study their products and current ads, then use AI video tools such as Seedance 2.5 to make each company one ad for free that you genuinely think people would watch.
After that, send an email with the pitch: "This ad is free. Run it. If it doesn’t make money, pretend I never contacted you. If it does make money, pay me $10,000 per month and I’ll keep making them for you."
Lee said winning just a handful of clients could be enough to build a high-margin AI advertising business.
He added that people can also use Meta Ad Library to find ads a company has been running over a long period, which may suggest they are working, then remake a variation with generative video and sell it to other competitors in the same sector.
The biggest AI arbitrage, Lee said, is the knowledge gap
Across all three models, Lee tied the opportunity to one central idea: the largest arbitrage in AI may not come from the models themselves, but from the gap between how fast AI capabilities are advancing and how slowly ordinary businesses understand them.
People who know the latest AI tools understand that a website may be built in a matter of hours. People who do not know that may still be willing to pay thousands or even tens of thousands of dollars to outsource the same work. People who understand AI video know a UGC ad may only require repeated prompting, while traditional companies may still follow the older process of hiring actors, camera crews, editors, and ad agencies.
That difference in understanding is the gold rush Lee was describing. In his view, the people most likely to make money from AI today are not necessarily the ones best at training models, but the ones who recognize early that AI can already do a specific job and are willing to sell that result to buyers who have not caught up yet.