AI Rebuilt Every YC W26 Startup. Should Founders Be Scared? | E2271

What It Covers

FeltSense CEO Marek Hazan used AI agents to rebuild every Y Combinator W26 batch startup, revealing that 10–20% were immediately replicable and ~30% were too hardware-dependent to clone. The episode also covers Bordie, an AI networking principal with 150,000 contacts, and a $1.8B GLP-1 telehealth company built by two brothers for $20.

Key Questions Answered

• •Startup defensibility audit: FeltSense's AI agents found that 10–20% of YC W26 startups were immediately replicable using modular, reusable components at low compute cost. Founders should proactively stress-test their own products by asking: can an AI agent rebuild this in an afternoon? If yes, the moat is technology alone — insufficient. Market access, proprietary data, and regulatory complexity are what actually create durable defensibility in 2026.
• •AI agent holding company model: Rather than selling AI cofounder tools to founders, FeltSense keeps all agents in-house, spinning up its own products across a scalable holding company structure — an AI-native version of IAC. The strategy targets the "long tail of entrepreneurship," starting with single- and dual-feature SaaS apps, aiming to run tens of thousands simultaneously. This avoids commoditization by Anthropic or OpenAI, who would simply replicate any externally sold cofounder tool.
• •AI networking principal vs. agent: Bordie, an AI built by Andrew D'Souza's company, operates as a "principled" connector rather than a simple agent — it can refuse user requests if an introduction would harm the network's reputation. With 150,000 contacts across founders, investors, and executives, Bordie monetizes via 20% contingency recruiting fees or $10,000/month retainers, while remaining free for most users. Creandum's entire partnership evaluated Bordie before ever speaking to the human founder.
• •Human-in-the-loop as legal protection: MedVie generated $400M in year-one sales and is tracking $1.8B in year two selling GLP-1s online, built by two people for $20. However, AI-generated ads allegedly used fake doctor names and fabricated before-and-after images, triggering an FDA investigation. The lesson: at scale, assign dedicated humans to review AI-generated outputs — especially ad creative — before publication. One compliance failure can result in criminal charges and permanent industry bans.
• •Data-first decision-making framework: When debating internal strategy — such as posting frequency on X — replace opinion-based discussions with structured AI-assisted data analysis. Pulling three months of X analytics into Claude and running a structured conversation produced charts and clear answers in one hour versus a full day of manual spreadsheet work. The operating principle: if data exists, analyze it first; if falling back on opinion, defer to the person with the most relevant lived experience.
• •Competitor PR as a competitive weapon: MedVie's New York Times exposure likely originated from a competitor tip, not organic journalism. Rival companies hire PR firms specifically to feed confidential tips to journalists about competitors cutting regulatory corners. Founders scaling fast in regulated industries — telehealth, fintech, food — should assume competitors are monitoring their ad libraries (searchable via Facebook's public Ad Library) and compliance posture, and proactively ensure all marketing claims meet FDA or FTC standards before reaching significant revenue.

Key Topics