On the fourth floor of 140 New Montgomery Street, super{set} is attempting something that sounds more mechanical than it feels: turning company creation into a repeatable craft. The studio does not simply wire money to founders and wait for quarterly updates. It begins at the whiteboard, helps define the problem, recruits a product-minded co-founder, supplies an early technical team and stays close while the new business searches for customers. The raw material is usually data. The intended result is enterprise software that does an unglamorous job well enough to become hard to remove.
That formulation separates super{set} from two crowded groups. It is not a conventional venture fund choosing among finished pitch decks, though it manages venture funds. It is not an accelerator pushing a large class toward demo day, either. The studio originates many of its own ideas, tests them with prospective users, and concentrates on only a few new companies each year. By 2024, it said it had founded, funded and scaled 16 startups. The portfolio stretches from privacy controls and automated software testing to health intelligence, advertising and autonomous customer acquisition.
The studio is the first co-founder
Tom Chavez and Vivek Vaidya started super{set} after a long run building data businesses together. Chavez co-founded Rapt, where Vaidya served as chief technology officer; Microsoft acquired the company. The pair later co-founded Krux, which Salesforce bought in 2016. When super{set} launched publicly in 2019 with a $65 million fund, the premise was that the bruises from those companies could become useful infrastructure for the next ones.
The studio's process starts before a product exists. A team defines a market problem, estimates the opportunity, maps likely use cases and speaks with prospective customers. Instead of a venture capitalist's investment memo, super{set} writes what it calls a solution memo. The document is meant to be a shared blueprint: What hurts? Who will pay to fix it? What proprietary data could improve the product over time? Is there enough room for a company rather than a feature?
Once an idea survives, the studio recruits a product-centric co-founder. Shared specialists contribute engineering, hiring, finance, legal, marketing and sales help. That support is heaviest at the beginning and becomes lighter as the company develops its own leadership. Crucially, there is no fixed graduation date. The arrangement asks founders to trade more equity than they might give a passive seed investor for a larger founding bench and fewer early administrative distractions.
“Partnering with super{set} isn't like having an investor; it's like starting day one with an entire team of seasoned co-founders.”Gal Vered, co-founder of Checksum
The model is shared. The data is not.
The fashionable AI question is which foundation model will dominate. super{set} is asking a less theatrical one: what useful appliance can be built on top? Its partners distinguish the science of AI from the engineering of AI. The first advances general capability. The second connects that capability to messy workflows, permissions, histories and outcomes inside a real business.
That distinction explains an otherwise eclectic portfolio. Ketch manages privacy and data governance. Checksum automates full-stack software tests. Headlamp Health helps providers assemble patient history and clinical insights. PointHealth connects healthcare organizations through a secure data vault. Rembrand places advertising inside video experiences. Parallel Distribution works on autonomous customer acquisition. The industries vary, but the mechanism repeats: capture difficult data, organize it, apply intelligence and turn the result into an action someone values.
The current version of this thinking appears in super{set}'s argument for “task-absorbing” software. Traditional software often produces alerts, dashboards and forms, leaving people to move information between systems. Agentic products can act: update the customer record, run the test, assemble the context or complete the follow-up. For super{set}, the durable advantage is not access to the same large model everyone else can rent. It is the domain context, proprietary feedback and permission structure that make an agent trustworthy inside a particular job.
Money with sleeves rolled up
super{set}'s business model blends fund economics with founder labor. External limited partners, family offices, technology investors and the founding partners supply capital. The studio invests in businesses it helps create, receives meaningful ownership and participates in the value of future financing or exits. Chavez and Vaidya have also invested some of their own proceeds, aligning their outcome with their funds.
The most revealing number may be the smallest. In 2024, the studio told TechCrunch it expected to reduce its pace to two or three new companies a year, from four or five earlier on. A factory usually wins through throughput. super{set}'s claimed edge requires the opposite: enough concentration for experienced operators to notice a bad product sequence, make the first introductions or help recruit the leader who changes the trajectory.
Its “Hive” is designed to multiply that attention. Portfolio executives compare notes, reuse management habits and act as design partners for one another. Kapstan, a cloud-management company in the portfolio, found an early user in Headlamp Health. The health startup could deploy without immediately hiring a full-time DevSecOps engineer; Kapstan got a live environment in which to refine its product. That is the studio flywheel in miniature: one company's inconvenience becomes another company's market evidence.
Proof, pressure and the founder's bargain
The clearest proof point arrived in January 2024, when LiveRamp agreed to acquire Habu for approximately $200 million in cash and stock. Habu's data clean-room software helped companies collaborate across decentralized data without surrendering ownership or privacy. Early customers included Disney, L'Oreal and PepsiCo. The company grew from lessons the founders had encountered at Krux, which is precisely the kind of carried-forward pattern recognition the studio promises.
Spectrum Labs supplied another exit when ActiveFence acquired it in 2023. Its AI systems detected toxic content for online communities including Riot Games, Grindr and The Meet Group. These outcomes show that the studio can help create businesses acquired by strategic buyers. They do not settle the broader question of repeatability. Venture returns are lumpy, young portfolios take years to mature, and a shared playbook cannot erase market timing or founder judgment.
The interesting experiment is not whether one studio-built company can work. It is whether the lesson from that company changes the odds for the next one.
Founders must evaluate a corresponding bargain. Building with super{set} can remove the lonely scramble for seed money, first hires and basic operations. It also means starting with a consequential institutional co-founder and negotiating ownership from day one. Compared with joining a Series A company, the recruit gets more influence. Compared with starting alone, the recruit gets less blank-sheet autonomy. The right choice depends as much on temperament as on terms.
Where super{set} sits now
In the market, super{set} occupies the narrow strip between seed investor, accelerator and operating company. Atomic, Science, Pioneer Square Labs and High Alpha offer versions of the venture-studio model; independent founders and conventional venture firms remain the default alternatives. super{set}'s differentiation is its deliberately tight lane: B2B software, data engineering, applied AI and complex enterprise workflows, with an operating relationship intended to last beyond an accelerator cohort.
The studio is also becoming more public about its thesis. Through events, videos and essays, its partners now discuss knowledge graphs, agentic systems, go-to-market coordination and the cost of fragmented software. Programs such as Signal invite entrepreneurs and technical leaders to bring domain insight into the company-formation process. The office at 140 New Montgomery doubles as a meeting place for that community.
There is a pleasing stubbornness to the wager. While much of the AI market races toward general intelligence, super{set} keeps returning to narrow software with a buyer, a workflow and a proprietary trail of data. Privacy controls are not cinematic. Test automation rarely trends. A secure clinical history will not write a sonnet. But enterprises pay for systems that reduce cost, manage risk and finish work. If the AI boom is going to become ordinary business infrastructure, someone has to build the ordinary parts.