After 50+ founder interviews, six AI tools keep recurring, and context-feeding beats clever prompting for making money.
Yang Zhao, CEO of Opus Clip, says treat AI as a thinking partner, not just a tool to design a poster or vibe code a prototype. When you struggle with understanding users, managing teams or figuring out pricing, run the decision through Gemini or ChatGPT with as much context as possible and more than 20 rounds of back and forth.
Once a month he asks ChatGPT what his major decisions of the past month were and asks for comments and feedback. He shares everything beforehand: sometimes he talks it through at night, sometimes he forwards screenshots of group discussions or drops the link to a PRD or spec. It forces him to document.
The entire team now works in Claude projects. The same producers as a year ago put out twice the content every month, which doubled revenue. There is a project per social channel, and the YouTube one has her voice, past episode performance connected to Notion, the topics her audience engages with and her interview style.
At Workera, a big Anthropic shop where engineers use Claude Code Max, skills are files defining how the company does something: how they recruit, brand guidelines, the font, how they speak, the colour palettes. Instead of asking marketing to review a website, the engineer asks the LLM to verify copywriting and palette.
She has 36 proactive workflows with 28 master agents, and each spins up about two sub-agents, call it 50, so somewhere around 100 agents in total. Her focus is what AI can do without her kicking it off: scheduled tasks that run while she sleeps, walks or does something else.
Most come in as emails routed into folders. Every Friday morning she gets a recap of urgent emails she has not answered, ranked by urgency, with drafted replies, the ability to delegate to her team and reminders, scraped from five days of Gmail. Every morning a briefing follows.
Three documents do the work. An anti-AI writing style file: how to never sound like AI, no generic AI phrasing, no filler, no cliches. A voice profile: how she actually writes, tone, rhythm, openings, endings, favourite vocabulary. And a fact dossier of verified context about her, her work and her audience.
Two people who knew neither chess nor programming started it. They researched what existed for learning chess, found it not great, downloaded Cursor and vibe coded puzzles, trained the AI on an online puzzle database when it struggled, then made mobile prototypes until the CEO approved. Chess now has 7 million daily active users.
On the 15th of every month it pulls from QuickBooks and sends a full picture: margin, projected tax owed, tax strategies to lower the bill this month, comparison to last year and what is eating profit, replacing an accountant charging by the hour. It also executes a Claude-built investment strategy, tracking dips for dollar cost averaging.
He pits the models against each other. He starts with Gemini, which feels like an American scientist, then asks DeepSeek what is missing, and DeepSeek says it is too American and names what was left out. He then gives it to ChatGPT to write better, and hands that back to Gemini or Grok.
“They are actually a very, very senior, very omnipotent thinking partner.”
“Then it turns into the advisor that catches mistakes before they cost you money.”
“And so what I do is I pit them against each other.”
“I believe that today I'm borrowing maybe 80 IQ points from my AIs.”
“I have 36 proactive workflows with 28 master agents”
“And again, these people did not know any chess.”
“Double Your Income With AI in 3 Months (Here's the Stack).” https://www.youtube.com/watch?v=zL2PIa72gJ4. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/double-your-income-with-ai-in-3-months-here-s-the-stack.html.