Can AI Build a Startup in 72 Hours? Musk's Team Will Livestream the Whole Thing

Can three people and a stack of AI agents build an actual startup in 72 hours, with the cameras rolling the entire time?

That's the test Elon Musk's AI company is running from September 15 to 17. Three staffers, Matt Palmer, Lauren Tan and Roshan Sadanani, will livestream every product, engineering and business decision as they build a company from zero. No name. No product. No plan. They're supposed to pick all three on camera, in real time.

The question worth asking: if agents can carry that much of the load, what's actually left for the founders?

What's on the Clock

Three days. That's the whole budget. For context, a typical seed-stage team spends weeks on discovery and months on a first usable build. Y Combinator hands founders three months and a small check, and even that feels brutal to most of them. This crew gets a long weekend, plus whatever internal tooling and compute they can pull in.

The output is undefined too. No name, no market, no roadmap. They'll choose those live. Which is either the most honest way to test an agent stack, or the least reproducible experiment anyone could design. I lean toward the second one.

Why It Matters Past the Stunt

Agentic AI has been the loudest narrative in crypto for over a year, and the reasoning is straightforward. Traders keep pricing in a future where software agents transact, negotiate and pay each other onchain without a human clicking approve. Tokens tied to that thesis have moved hard on far less than a three-day livestream.

But here's the piece most demos skip. Nobody has publicly shown a full company, from naming through shipping, run mostly by agents under deadline pressure. Weekend hackathons have existed for nearly two decades, and Startup Weekend style events have produced real companies, though rarely in 72 hours. What's new isn't the format. It's who, or what, does the work.

If agents handle naming, design, code and go-to-market, the founder becomes an editor. That's a genuine shift in how early-stage companies get built, and it shapes where venture dollars and developer attention flow next.

Who's Buying It

Color me skeptical, but a demo run by the company that makes the tools, staffed by its own employees, on its own infrastructure, isn't a neutral test. It's a showcase with a livestream attached. That doesn't make it worthless. It just means the results won't tell you much about what happens when a solo founder with a $20 monthly budget tries the same thing on a Tuesday night.

Skeptics will argue the interesting calls get made off camera, in a repo, by humans who happen to be fast. Proponents will argue the speed itself is the point. Both can be true inside the same 72 hours.

Investors are watching for a different reason. Any sign that agent orchestration holds up at this scale feeds the case for agent-native payments, wallets and marketplaces, which is where a lot of crypto capital has been parked for months now.

What to Watch Next

Mark the dates: September 15 through 17. Then track four things. Does the code get published, or does the stream end with a demo and a promise? Does the startup have a single real user by hour 72? Do they disclose which models and tools carried the load? And does anything survive past day three, when the cameras go dark and the team returns to their day jobs?

If a working product with actual users comes out the other side, that's a meaningful data point for every builder weighing an agent-first workflow. If it's a polished three-day advertisement, the tooling still probably matters, just less than the hype suggests.

The stream starts September 15. I'll be watching the repo, not the countdown clock.