The Portfolio Machine Thesis: What Happens to Founders When VCs Can Spin Up a Thousand MVPs

If AI can take a product from trend detection to working MVP for a few thousand dollars in tokens, capital allocators no longer need to bet on one founder with one idea. They can run a thousand concurrent experiments and kill the 950 that produce no revenue. That is the thought experiment Camilla Castro runs through in this 82-second video, filmed casually from her couch under a static caption that frames the whole argument: "From VC firms to portfolio machines (The end of founders?)". She lands on a more measured answer than the hook suggests. Founders are not dead, but the job is shifting from building one idea to building a portfolio of ideas.

The Core Argument: Capital to Founders Versus Capital to Machines

Castro's setup is a cost claim. Assuming the right tech stack, AI can now spot a trend, define the product requirements, write the strategy, and build the actual product for close to nothing. Her description puts a number on it: a few thousand dollars in tokens plus some ad spend per product. At that cost per experiment, the traditional VC model of allocating capital to a founder with a single idea looks inefficient. Her alternative framing is a portfolio machine. Spin up a thousand MVPs, run them as concurrent experiments, kill the 950 that produce no revenue, and scale the 50 that do. As she puts it in the caption, "That's not venture capital. That's a portfolio machine."

The Real-World Test Case: Audos

Castro says she was curious whether any firm actually operates this way and found one startup studio in New York City trying to launch a hundred thousand companies in a year. That describes Audos, founded by Henrik Werdelin (who previously built the Prehype studio behind Barkbox) and Nicholas Thorne. TechCrunch covered the company in June 2025. Audos gives non-technical founders AI tools to build products through natural language, plus up to $25,000 in funding and distribution help through paid social. Instead of equity, it takes a 15% revenue share, and it raised an $11.5 million seed round led by True Ventures. The model is close to Castro's thought experiment with one difference: Audos still puts a human founder at the center of each company rather than running the experiments itself.

Where AI Stops

Castro lists the concrete blockers to a fully founderless model. AI cannot sign a contract, cannot get sued, and cannot open a bank account. Legal and financial personhood still requires a human. She also notes that AI does not entirely know which problems are worth solving, though she concedes the models are getting good at identifying opportunities. These limits explain why the portfolio machine still needs people in the loop, even if the people look more like operators of an experiment pipeline than traditional founders.

The 2AM to 10AM Proof Point

The idea got personal for her that morning. She spun up a mini idea she had at 2am and by 10am it was already working. That eight-hour idea-to-working-product loop is the individual-scale version of the portfolio machine argument. Her conclusion follows from it: the founder job is changing. You no longer have to be a builder of one idea. You can be a builder of a portfolio of ideas. She closes by saying she wants to test this herself and invites viewers to follow along to see what she builds next.

Presentation Notes

The video is a single talking-head shot. Castro is reclined on a green couch with a dog visible at the edge of frame, wearing glasses and speaking directly to camera. The only graphics are the persistent title card at the top and yellow-highlight karaoke captions at the bottom. There are no demos, charts, or screen recordings. The production choice fits the content: it is presented as a thinking-out-loud thought exercise, and she labels it as such in the video.

Key Takeaways

• If an MVP costs a few thousand dollars in tokens, the rational capital strategy shifts from backing one founder to running many concurrent experiments and killing the losers fast.
• The model she describes exists in early form: Audos in New York aims to launch 100,000 companies a year, backed by an $11.5M seed from True Ventures, using a 15% revenue share instead of equity.
• Hard blockers to a founderless model remain legal and financial: AI cannot sign contracts, be sued, or open bank accounts.
• Problem selection is still a human edge, though she concedes models are getting good at spotting opportunities.
• Her own test run went from a 2am idea to a working product by 10am, which is the individual version of the same thesis.
• Her conclusion: founders are not obsolete, but the role is shifting from one-idea builder to portfolio builder.

Resources

• Original TikTok by @camillasilvacastro
• Audos - the New York startup studio referenced in the video
• TechCrunch: This AI-powered startup studio plans to launch 100,000 companies a year — really