Founders' AI Stack: 2x Revenue via Thinking Partners & Agents
From 50+ founder interviews: Treat ChatGPT as a thinking partner with deep context (20+ rounds), use Claude projects for team workflows (doubled output/revenue), deploy 100-agent systems for proactive automation—tools that actually move the needle on income.
AI as Thinking Partner: Feed Context, Iterate Deeply
Yang Xiao, CEO of Opus Clip (0 to 50M users, $215M valuation in 2.5 years), treats ChatGPT not as a quick query tool but as a "senior thinking partner." Instead of one-line questions, he dumps full context—screenshots, PRDs, group discussions—and runs 20+ rounds of back-and-forth. Monthly, he reviews decisions with it: "What are my major decisions in the past month? Give feedback." This catches regrets before they scale, replacing coaches or mentors. Tradeoff: Requires forcing documentation habits; no magic without input volume. Result: Enlightened decisions on users, teams, pricing. Speaker's twist: ChatGPT for emotional support ("always on your side"), switch to Claude/Gemini/Perplexity when needing tough love.
"The number one AI skill should actually go for first principle... treat AI as your thinking partner... throw as many context as possible um and also you know do like more than 20 rounds of back and forth um communications. you will be mindblowingly enlightened." — Yang Xiao, explaining why a $215M CEO still defaults to ChatGPT daily.
Multi-Model Pitting: Borrow 80 IQ Points Without Outsourcing Judgment
Mo Gawdat (ex-Google X CBO) rejects single-model monopoly: Start with Gemini ("scientist, American bias"), critique with DeepSeek ("too American, missing politics/motivation"), polish with ChatGPT ("writes elegantly, California-nice"). Repeat until truth emerges. Why? AI appears authoritative but folds under pushback—users must verify. He compares to engineering school: Calculators halved solve time; smart students doubled-checked. Tradeoff: Time-intensive upfront, but amplifies human intelligence on info-crunching/search. Outcome: "Borrowing maybe 80 IQ points from my AIs... AI IQ is exponential." Business idea: Build a comparator chat.
"AI is going to make you dumb if you outsource your problem solving to AI. AI is going to make you the smartest you've ever been. If you take the parts that are not natural to the human brain... but get the AI to do the work so that you do the intelligence." — Mo Gawdat, on using AI to 2x problem-solving, not halve effort.
Claude Projects: Embed Team Knowledge to 2x Output
Post-interview with Ken Katan-Fouch (Stanford AI co-founder), speaker rebuilt team ops around Claude "projects"—persistent workspaces with "skills" (files defining processes). Examples: Brand guidelines (fonts, voice, palettes), recruitment playbooks. Engineers query for compliance, slashing marketing handoffs. Speaker's setup: Per-social Claude project (YouTube/LinkedIn/newsletter) ingesting Notion DB—past performance, audience topics, interview style. Result: Same team doubled monthly content, doubled revenue in weeks. Claude even built GEO (generative engine optimization) strategy sans specialist. Tradeoff: Maintenance overhead (update files), but frees humans for strategy. Non-obvious: Still hire outsiders for blind spots.
"Before, if an engineer wanted to build a website, they would have to call the marketing team... Today... the engineer just asks the LLM, can you just verify... And you gain actually so much speed." — Ken Katan-Fouch, on Anthropic's internal Claude use at Workera.
100-Agent Systems: Proactive Workflows Replace Manual Kicks
Allie Miller (ex-Amazon AI leader) runs 36 proactive workflows via ~100 agents (28 master + sub-agents): Scheduled Gmail scrapes (Friday urgent email recap/drafts/delegations), morning briefings (industry news, events, meeting prep—runs overnight). Trigger with keywords (e.g., CEO meeting → auto-assets). 2-10x productivity vs. query-response. Platforms: Claude Co-work, Codeex. Tradeoff: Setup complexity, but automates "asking" friction. Speaker adopted for similar gains; most overlook scheduling.
"What can AI do that I don't have to kick off? ...every single Friday morning, I have a recap of all of the urgent emails... every morning I wake up, my AI agent has already been working for me for several hours." — Allie Miller, on her 100-agent system handling hours of delegated work.
Anti-Generic Files + Vibe Coding: Compete on Brand in Collapsed Cycles
Three files per AI/team member/platform: 1) Anti-AI style (no filler/clichés), 2) Voice profile (tone, vocab, examples), 3) Fact dossier (bio/audience). Transforms generic drafts to authentic. Speaker shares templates in newsletter. Trend: Vibe coding—describe in English, AI codes. Gary Vaynerchuk: "Hyper micro wealth" window for $5-50/mo apps (e.g., passport photos making $10k/mo). Duolingo CEO: Non-coders hit 7M DAU in 6 months. Why now? AI kills build moats; brand/audience understanding wins. Design.com demo: AI logos → full brand kit (sites, socials) in minutes, commercially safe.
"Learning to vibe code right now is a real window to build wealth and that window won't stay open forever... I would build an app that's $5 to $50 a month and... try to get customers." — Gary Vaynerchuk, on non-coders capturing long-tail demand before AI saturation.
Key Takeaways
- Dump full context (docs/screenshots) into ChatGPT + 20+ iterations: Builds advisor spotting decision flaws.
- Pit models (Gemini → DeepSeek → ChatGPT): Forces truth over bias; repeat for polish.
- Build Claude projects per channel/team: Embed voice/DB for 2x output without extra headcount.
- Deploy 36+ proactive agents: Schedule briefings/email recaps for overnight work.
- Upload 3 style files (anti-AI, voice, facts): Ends generic output; templates in speaker's newsletter.
- Vibe code micro-SaaS now: $5-50/mo niches persist despite AI commoditization.
- Use Design.com for instant brand kits: Logos → sites/socials; closes credibility gap fast.
- Document everything: AI memory unlocks monthly retrospectives on regrets.
"Most said that most people use AI to work less. The smart ones use it to earn more." — Speaker, contrasting lazy vs. leveraged AI use across 50 founders.