Zeroset – Akshat Kannan and William Zhang are building ‘Nebula’ to teach AI agents how businesses actually operate, securing backing from Gradient and 2048 Ventures.
The standard AI agent is a digital worker with a blind spot. While they can draft emails and summarize documents, they lack the institutional memory of how a specific company actually functions—a reality that keeps them from truly autonomous work.
Zeroset, a startup emerging from stealth today, is betting $5.2 million that it can bridge that gap. The company, founded by 18-year-old Akshat Kannan and 21-year-old William Zhang, is building models designed to give AI agents a constantly updated, living picture of a business’s operations.
“Models off the shelf have no understanding of the dynamics of how an enterprise actually works,” said Kannan, who left Stanford earlier this year to focus on the project. “They have no understanding of how the work gets done.”
For most companies. the reality of their internal workflow is scattered across a fragmented landscape of customer relationship management systems. ticketing platforms. and internal documentation. Current AI models are rarely trained on this proprietary data. Zeroset’s primary tool. Nebula. attempts to fix this by connecting directly to the software a company already uses—including Microsoft 365. SharePoint. Outlook. Teams. GitHub. and various workplace messaging tools.
By tracking how information and activity shift across these platforms. Nebula creates a roadmap of the sequences and decisions that define a workflow. The goal is to move beyond simple document retrieval. allowing an AI agent to see how similar cases were handled in the past and perform the work autonomously.
This shift in functionality is a response to the current limitations of enterprise AI, where information is isolated in disparate digital siloes. Zeroset’s platform turns those siloes into a cohesive history, providing the context required for high-level tasks.
The company. which launched in January after Zhang dropped out of the University of Texas last year. remains in a closed research preview. Rather than a broad public rollout, the founders are opting to expand access to customers progressively. Their initial target is clear: enterprises and AI-native firms deploying long-running agents in sectors like manufacturing. financial research. and supply-chain operations.
Their business model relies on a combination of a license fee and usage-based pricing, which will scale alongside the volume of data and agent activity flowing through Nebula.
Financial backing for the pre-seed round was co-led by Gradient—the venture capital fund founded by Google in 2017—and 2048 Ventures. with additional participation from Leblon Capital. The capital infusion arrives at a time when the startup is focused on aggressive growth; Zeroset currently employs five people and intends to use the new funding to bolster its team of researchers and systems engineers. fund early enterprise deployments. and cover the significant costs of model training.
These training costs are substantial. In one recent peak training session alone, the company spent approximately $100,000 in two weeks.
As Zeroset enters the space. it joins a growing market of companies attempting to build memory infrastructure for artificial intelligence. indirectly competing with firms like Zep and Mem0. For now. the focus remains on the quiet. behind-the-scenes work of teaching machines the internal rhythm of a business. one workflow at a time.
Zeroset AI agents Akshat Kannan William Zhang Gradient enterprise AI Nebula AI startup funding