In 1996, Bill Gross started Idealab. Over the next thirty years it launched 150 companies, took 45 of them public or sold them, and watched another 40 die. Seven became billion-dollar businesses. Bill is one of the great founders. He is also one of the great cautionary tales about studios.
The studio model works when it works, and it works in spasms.
The standard critique of studios has not changed since Idealab. Naval Ravikant put it best: a studio cannot attract the best founders, because the best founders will not give up forty percent of their company for an idea and a desk when a venture capitalist will give them money for fifteen. The studio is therefore left with the second-best founders, the second-best ideas, and the second-best outcomes. The seventy percent of studios that fail tend to fail this way.
We started Veronata in 2025 with this critique in mind. We knew the math.
The math is different now because of agents.
What broke the old studio
The historical studio failure mode is: one CEO, one company, one founder rationing attention across many things. A studio of seven products is a studio with seven CEOs, seventy employees, seven boards, seven payroll runs, and seven sets of all-hands meetings. The studio's general partner cannot pay attention to all of them. Talented founders detect the inattention and leave. The companies that remain are the ones whose founders had no better offers.
This is not a software problem. It is a coordination problem.
Bill Gross has talked about this in interviews. His most successful Idealab spinouts — Overture, Picasa — had founders who were going to start a company anyway and chose Idealab as a launchpad. The companies that started inside Idealab, with Idealab supplying the CEO from a rolodex, did much worse on average. The model selected for the wrong kind of founder.
What changes with agents
The agentic studio has one CEO and one set of agents. The agents do not need to be hired. They do not switch companies. They do not get diluted. They do not have meetings. They cost roughly three times what their human equivalent would cost per hour, and they work roughly three times as many hours. Conservatively, the throughput is about nine times what one human could produce alone.
We are not the first to make this observation. We are, as far as we can tell, the first to live in it.
Some specifics from the last twelve months:
- One human, seven products. Veronata's portfolio is StarSinger, Tuck, SnoreCam, PinkyBloom, PinkyBond, Quit336, and Tadula. There is one human. There is no team. There is no office of seven CEOs.
- Identification time is days, not months. Agents survey markets continuously. The interval from "spotted opportunity" to "shipped first version" is currently 6 to 10 weeks for an iOS app. The first iteration of Tuck took 7 weeks.
- Operations cost is logarithmic, not linear. Adding a seventh product to a studio of six adds roughly 15% to operating cost, not the 17% you would expect from one-seventh of the existing base. The agent stack — support, deploy, App Store paperwork, refund handling — is shared across all of them.
- Failure is cheap and informative. An agent-built product that fails dies within its first 90 days, at a total cost in the low five figures. Idealab's failures often took two years and high six figures.
What everyone else is missing
Y Combinator's most recent batch was its largest single bet on agents: 67 of the 144 Spring 2025 companies are building agents of one kind or another. Every single one of them is a single-product company. Not one of them is a studio.
This is the open lane.
A studio of consumer products operated by agents is not a new species. It is what holding companies have always wanted to be. The cost of running one used to make it untenable for anyone without a Buffett-grade insurance float. The cost of running one now is approximately seven Stripe accounts and one human's attention.
It is too early to know whether we are right. We have shipped seven things; none of them is a unicorn. Two are profitable. Three are not yet. Two are too young to tell. We will write about each launch's economics when we have a year's worth of operating data on each.
For now, the bet is on the table. If the studio model has been waiting for something, it has been waiting for this.