Pomegra
Zeroset, led by an 18-year-old and a 21-year-old, emerged from stealth with a $5.2M pre-seed round to give AI agents a live model of how a company works.
- Zeroset raised a $5.2M pre-seed co-led by Gradient and 2048 Ventures, with Leblon Capital participating.
- Its product, Nebula, tracks changes across enterprise tools so agents can see workflow history.
- CEO Akshat Kannan is 18 and co-founder William Zhang is 21. The company has five employees.
Lead
Zeroset, a US startup founded in January 2026, came out of stealth on October 6 with a $5.2 million pre-seed round. Gradient and 2048 Ventures co-led the round, and Leblon Capital also took part. The company builds models that give AI agents a constantly updated picture of how a business operates. Valuation was not disclosed.
What Does Zeroset Actually Build?
Zeroset builds Nebula, software that holds a running record of how work moves through a company. It connects to Microsoft 365, SharePoint, Outlook, Teams, GitHub and workplace messaging systems. It tracks changes in activity and information across them.
Most enterprise agents today retrieve a few relevant passages from a large document store. Nebula is designed to represent what changed, what decision followed, and how similar cases were handled. It turns each workflow into a typed, queryable trace, so an agent can draw on the sequence behind a task and not only the final documents.
The product competes for the same budget as retrieval tools and agent-memory layers. Zeroset's pitch is that retrieval answers "what does the file say" and leaves out "how did this get decided."
Who Are the Founders?
Akshat Kannan, the CEO, is 18 and dropped out of Stanford this year. William Zhang is 21 and left the University of Texas last year. They founded Zeroset in January.
Kannan frames the problem as a data gap. "Models off the shelf have no understanding of the dynamics of how an enterprise actually works," he told Business Insider. He said that knowledge sits scattered across customer relationship management systems, ticketing tools and internal documents.
The team is five people. That makes $5.2 million a large sum relative to headcount, and it will go mostly to hiring researchers and systems engineers, funding early enterprise deployments and paying for model training. Training costs peaked at roughly $100,000 over two weeks, according to the company.
How Will Zeroset Make Money?
Zeroset plans a hybrid of licensing fees and usage-based pricing that scales with the volume of data and agent activity. Initial target markets are manufacturing, supply-chain operations and financial research, across large enterprises and AI-native companies.
The product remains in a closed research preview. Access is expanding gradually, not through a broad launch. No customer names or revenue figures have been disclosed, so the round rests on the founders' technical thesis and not on commercial traction.
Why Does This Round Matter?
The round matters because it shows investors will fund very young teams on a specific bet: that agent reliability depends on organizational context and not only on larger base models. A pre-seed of this size for a team with no public customers is a sign of how much capital is looking for an angle on enterprise agents.
The skeptical reading is straightforward. Large platform vendors already own the data in Microsoft 365 and similar suites, and they can build comparable state tracking into their own agent products. Zeroset has to show that a neutral layer across tools outperforms what incumbents ship natively. Access permissions and privacy controls will also decide whether companies let a third party record how their staff work.
What Comes Next for Zeroset?
Next for Zeroset is hiring and the first paid enterprise deployments. The company has said it will use the round for researchers, systems engineers and training compute, which points to model quality as the near-term priority over sales.
If early deployments show agents completing multi-step tasks with fewer errors, a seed round would likely follow within a year or so. If the incumbents' native tooling closes the gap, Zeroset would need to narrow to the verticals it has named, such as supply-chain operations and financial research, where its trace-based approach has the clearest edge.
Outlook
Zeroset enters the market with $5.2 million, five employees, two founders under 22 and a product still in research preview. The thesis is credible and the investors are specialist AI backers, but customers, pricing power and defensibility against platform vendors remain unproven. The next two quarters of enterprise pilots will show whether Nebula's workflow traces improve agent performance enough to justify a separate product.
